{ "summary": { "attemptedTrials": 128, "terminalTrials": 126, "trialsWithOffloadPlans": 128, "offloadPlanEvents": 128, "distinctPlacement": [ { "gpu": 39, "total": 64, "cpu": 25 } ], "launches": 139, "contexts": [ 65536 ], "slots": [ 1 ], "kv": [ "f16" ] }, "note": "Actual task daemon logs only; excludes admission/preflight probes. Placement frequency is per attempted trial and logged plan, not time-weighted utilization or all-CPU execution.", "trials": [ { "cellId": "r1-saluki-cuda-core-tictactoe", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-core-tictactoe/tictactoe/tictactoe-underdog-saluki-27b-iq2-mix-2026-11-04T00-05-25-964Z-gpek", "terminal": true, "logSha256": "d9c8e70e8e842930ff7b60853cfe24433b5cd03089ef81197e6d7bc89cd35089", "plans": [ "2026-11-04T00:05:32.874Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 36460, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-core-petshop", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-core-petshop/petshop/petshop-underdog-saluki-27b-iq2-mix-2026-11-04T00-16-26-050Z-1i5w", "terminal": true, "logSha256": "2739852e52dcf1225f4f36a076f1331a976e044221ab76385e69553a560dbda7", "plans": [ "2026-11-04T00:16:33.285Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 13912, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" }, { "pid": 2852, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-core-tankcombat", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-core-tankcombat/tankcombat/tankcombat-underdog-saluki-27b-iq2-mix-2026-11-04T01-40-41-736Z-5r29", "terminal": true, "logSha256": "6b85593210bda8fae2ca90cc72348b008e373c1c9866b8ce97cbc8bf75dc0e44", "plans": [ "2026-11-04T01:40:51.320Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 36220, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-core-schema-migration", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-core-schema-migration/schema-migration/schema-migration-underdog-saluki-27b-iq2-mix-2026-11-04T02-13-44-132Z-3p7z", "terminal": true, "logSha256": "f92121e7f1ed292200899f493b79a76e106a609e8de56ab0a6df6e3ee776d536", "plans": [ "2026-11-04T02:13:55.177Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 12020, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-core-failing-tests-spec", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-core-failing-tests-spec/failing-tests-spec/failing-tests-spec-underdog-saluki-27b-iq2-mix-2026-11-04T02-44-03-484Z-60uo", "terminal": true, "logSha256": "939ab1fdc2c127187bc12fa36e3acbaee11c32986097fdd299f2c62666dabacf", "plans": [ "2026-11-04T02:44:13.048Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 37656, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-core-symptom-debug", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-core-symptom-debug/symptom-debug/symptom-debug-underdog-saluki-27b-iq2-mix-2026-11-04T03-05-54-026Z-2pi1", "terminal": true, "logSha256": "908f8e729a1eba10c60b32273cf878c0284a4186be16933712db12f5573ea539", "plans": [ "2026-11-04T03:06:03.604Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 51808, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-core-data-wrangle", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-core-data-wrangle/data-wrangle/data-wrangle-underdog-saluki-27b-iq2-mix-2026-11-04T03-15-59-497Z-0stw", "terminal": true, "logSha256": "ba0d86608e525cbe9a59bbf1a9b3d5531e98a334c95e7c12baa7116090bad03c", "plans": [ "2026-11-04T03:16:08.695Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 10420, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-core-incident-postmortem", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-core-incident-postmortem/incident-postmortem/incident-postmortem-underdog-saluki-27b-iq2-mix-2026-11-04T04-15-17-173Z-0cek", "terminal": true, "logSha256": "5bbe95f58ed499ffbdab0f0b7bcf0aa6fca2a94010f3f6ff5a988e532bc9d95b", "plans": [ "2026-11-04T04:15:27.787Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 48888, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-core-ops-runbook-anomaly", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-core-ops-runbook-anomaly/ops-runbook-anomaly/ops-runbook-anomaly-underdog-saluki-27b-iq2-mix-2026-11-04T04-51-33-586Z-3pp9", "terminal": true, "logSha256": "3bc9b9df61939de2262e96f0fb8e8fdb2748501d739516527a9ed232203a681e", "plans": [ "2026-11-04T04:51:42.704Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 14360, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-core-plan-and-estimate", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-core-plan-and-estimate/plan-and-estimate/plan-and-estimate-underdog-saluki-27b-iq2-mix-2026-11-04T05-06-14-544Z-4tg9", "terminal": true, "logSha256": "da49f109a9a81f8c31fd79e876d47cef554967c4aa4b127fd9f41513ff1eb471", "plans": [ "2026-11-04T05:06:25.077Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 50676, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-core-conflict-synthesis", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-core-conflict-synthesis/conflict-synthesis/conflict-synthesis-underdog-saluki-27b-iq2-mix-2026-11-04T05-17-18-555Z-4jcd", "terminal": true, "logSha256": "c678d0466b100b341c7d711542f146ac50844b6bf97d1ad212cab7418d891ff9", "plans": [ "2026-11-04T05:17:28.336Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 6636, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-productivity-constrained-comms", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-productivity-constrained-comms/constrained-comms/constrained-comms-underdog-saluki-27b-iq2-mix-2026-11-04T06-15-01-377Z-945b", "terminal": true, "logSha256": "988f191dc4d9056f8e26738fb1abcd187dd6442330400d1a3c532a11f8e86bce", "plans": [ "2026-11-04T06:15:11.101Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 17332, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-productivity-records-intake", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-productivity-records-intake/records-intake/records-intake-underdog-saluki-27b-iq2-mix-2026-11-04T06-25-47-184Z-1jry", "terminal": true, "logSha256": "aeb9716820cf445c842b0781fcf988d696f5895fe070b25268a761768dd96ec1", "plans": [ "2026-11-04T06:25:56.914Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 7752, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-productivity-plan-and-estimate", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-productivity-plan-and-estimate/plan-and-estimate/plan-and-estimate-underdog-saluki-27b-iq2-mix-2026-11-04T07-20-21-294Z-4kek", "terminal": true, "logSha256": "68d41b6c80293fb56d50aa00654d19e50565e3bacaa2b5095de558c945045265", "plans": [ "2026-11-04T07:20:31.060Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 47944, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-productivity-meeting-followup", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-productivity-meeting-followup/meeting-followup/meeting-followup-underdog-saluki-27b-iq2-mix-2026-11-04T07-31-51-777Z-429c", "terminal": true, "logSha256": "eef850ff0cbf998ee5dec0c68e73d3a969c66763cdb55474912b1adf84171f4f", "plans": [ "2026-11-04T07:32:01.550Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 34476, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-productivity-craftbook-spreadsheet-model", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-productivity-craftbook-spreadsheet-model/craftbook-spreadsheet-model/craftbook-spreadsheet-model-underdog-saluki-27b-iq2-mix-2026-11-04T07-47-54-584Z-8tqz", "terminal": true, "logSha256": "d362fc5c5a934fdd375f91e9cac16898a24e979e349c410f45b2c6e62006f6fd", "plans": [ "2026-11-04T07:48:09.512Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 31060, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-productivity-conflict-synthesis", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-productivity-conflict-synthesis/conflict-synthesis/conflict-synthesis-underdog-saluki-27b-iq2-mix-2026-11-04T09-08-33-934Z-i0ew", "terminal": true, "logSha256": "f062547058a5cc2c5ba1ca1cddab3197c845671cac31a2b513b005c13799667e", "plans": [ "2026-11-04T09:08:43.695Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 18392, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-productivity-docblocks-theme-roundtrip", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-productivity-docblocks-theme-roundtrip/docblocks-theme-roundtrip/docblocks-theme-roundtrip-underdog-saluki-27b-iq2-mix-2026-11-04T09-39-57-681Z-tkrr", "terminal": true, "logSha256": "2df6f9af948b34b73624b95b85dbc1a813c76a39c1780255a4cded38fae1ee47", "plans": [ "2026-11-04T09:40:07.452Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 38512, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" }, { "pid": 42332, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-productivity-craftbook-powerpoint-deck", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-productivity-craftbook-powerpoint-deck/craftbook-powerpoint-deck/craftbook-powerpoint-deck-underdog-saluki-27b-iq2-mix-2026-11-04T10-01-30-180Z-gc31", "terminal": true, "logSha256": "028150d246d92a98d7de276bfad1e0ebfa47b29a027dc3ac73c05aa005cb08c5", "plans": [ "2026-11-04T10:01:44.828Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 38080, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-productivity-wikipedia-research-brief", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-productivity-wikipedia-research-brief/wikipedia-research-brief/wikipedia-research-brief-underdog-saluki-27b-iq2-mix-2026-11-04T11-17-39-776Z-5ncd", "terminal": true, "logSha256": "b02df4f73265c7af93c4eb9c7d2669f31bf095a4a2cbd5878a4d61fd9b815361", "plans": [ "2026-11-04T11:17:49.614Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 14548, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-productivity-craftbook-week-plan", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-productivity-craftbook-week-plan/craftbook-week-plan/craftbook-week-plan-underdog-saluki-27b-iq2-mix-2026-11-04T12-04-36-260Z-bkxi", "terminal": true, "logSha256": "f2cefc1d04cbb0cedbf47949d34bd53dcdb4a31a8585c4f78619bd0848e8746d", "plans": [ "2026-11-04T12:04:49.282Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 18728, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-productivity-craftbook-ab-test-readout", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-productivity-craftbook-ab-test-readout/craftbook-ab-test-readout/craftbook-ab-test-readout-underdog-saluki-27b-iq2-mix-2026-11-04T13-40-24-956Z-dbhi", "terminal": true, "logSha256": "6b48ef36b114e63ec09ce2a50e1ea6a6ca80f919a48100cdf4d2e29f70500701", "plans": [ "2026-11-04T13:40:37.991Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 3984, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-productivity-craftbook-annotated-bibliography", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-productivity-craftbook-annotated-bibliography/craftbook-annotated-bibliography/craftbook-annotated-bibliography-underdog-saluki-27b-iq2-mix-2026-11-04T15-04-12-085Z-9axo", "terminal": true, "logSha256": "7a43785c39edf4b43be7141c3a1e70e08255ef4e16b791925fd9cd49b8aa5f6c", "plans": [ "2026-11-04T15:04:24.645Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 50144, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-productivity-craftbook-research-to-document", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-productivity-craftbook-research-to-document/craftbook-research-to-document/craftbook-research-to-document-underdog-saluki-27b-iq2-mix-2026-11-04T15-28-35-768Z-krj5", "terminal": true, "logSha256": "1e6a54fd34684c3ac4aa5d5facc28dd9b4e9b3167afbad7bc6b6a763e0e22eaa", "plans": [ "2026-11-04T15:28:48.402Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 50160, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-developer-fix-squisq-bugs", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-developer-fix-squisq-bugs/fix-squisq-bugs/fix-squisq-bugs-underdog-saluki-27b-iq2-mix-2026-11-04T15-51-11-509Z-0c2q", "terminal": true, "logSha256": "e493bc2202368e8cf0dd414960dba73316251282a5cdc030086f80cf852f494f", "plans": [ "2026-11-04T15:51:21.383Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 37156, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" }, { "pid": 45432, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-developer-dev-craftbook-routing", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-developer-dev-craftbook-routing/dev-craftbook-routing/dev-craftbook-routing-underdog-saluki-27b-iq2-mix-2026-11-04T16-08-27-562Z-yxbe", "terminal": true, "logSha256": "08c5dd0954c4f9272d3d50887f8e7c8c19985e50a05153ceb55dd9262cda42ad", "plans": [ "2026-11-04T16:08:35.901Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 43820, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-developer-interface-contract", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-developer-interface-contract/interface-contract/interface-contract-underdog-saluki-27b-iq2-mix-2026-11-04T16-12-00-765Z-8y6i", "terminal": true, "logSha256": "811490711af0d93d4b8f930c872ac02fc240705128fecf43e0438a8d8cffb791", "plans": [ "2026-11-04T16:12:08.165Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 31504, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-developer-craftbook-code-review", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-developer-craftbook-code-review/craftbook-code-review/craftbook-code-review-underdog-saluki-27b-iq2-mix-2026-11-04T16-17-39-748Z-7wv0", "terminal": true, "logSha256": "e067f702a33d5865370e1742df30f52f424f5c4669bd4e8a7a76ff0ede3fec8e", "plans": [ "2026-11-04T16:17:51.174Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 29868, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-developer-craftbook-api-contract-review", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-developer-craftbook-api-contract-review/craftbook-api-contract-review/craftbook-api-contract-review-underdog-saluki-27b-iq2-mix-2026-11-04T16-46-58-460Z-lbm2", "terminal": true, "logSha256": "18e00dc492a66a007d6f4336daed7e986d712cacc2c49412b554a55d6385edf8", "plans": [ "2026-11-04T16:47:10.170Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 43468, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-developer-craftbook-deep-security-review", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-developer-craftbook-deep-security-review/craftbook-deep-security-review/craftbook-deep-security-review-underdog-saluki-27b-iq2-mix-2026-11-04T17-45-22-802Z-hot9", "terminal": true, "logSha256": "95a3b32c73529e91c950d4168b60fcdae7ce70a47428516a3c275c608e585e13", "plans": [ "2026-11-04T17:45:32.729Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 35192, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-developer-large-pr-review", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-developer-large-pr-review/large-pr-review/large-pr-review-underdog-saluki-27b-iq2-mix-2026-11-04T18-07-39-731Z-bpdj", "terminal": true, "logSha256": "c87b37e2450c5434f095150a8e1a0580cfa3cad30bb855321a457b66aa0c4b6d", "plans": [ "2026-11-04T18:07:49.620Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 48192, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-developer-craftbook-codemod-sweep", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-developer-craftbook-codemod-sweep/craftbook-codemod-sweep/craftbook-codemod-sweep-underdog-saluki-27b-iq2-mix-2026-11-04T19-37-25-765Z-tdx6", "terminal": true, "logSha256": "e9e20b2370e6f4ed7d0b57033367007b3e7231931360a55c9ec9fb6be95dab09", "plans": [ "2026-11-04T19:37:37.255Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 37500, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-developer-craftbook-refactor-module", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-developer-craftbook-refactor-module/craftbook-refactor-module/craftbook-refactor-module-underdog-saluki-27b-iq2-mix-2026-11-04T20-08-16-625Z-9gn5", "terminal": true, "logSha256": "9bba3a4217e79f150fb68ba97f9dc190739ea7eba22a3837bd9715baf73a0909", "plans": [ "2026-11-04T20:08:28.219Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 12220, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-developer-craftbook-bug-fix-tdd", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-developer-craftbook-bug-fix-tdd/craftbook-bug-fix-tdd/craftbook-bug-fix-tdd-underdog-saluki-27b-iq2-mix-2026-11-04T21-11-12-350Z-8q4h", "terminal": true, "logSha256": "82861bb185a1afeb8e98b24e88da46c4e4b2dee81d1f6e55961a669da0066fc5", "plans": [ "2026-11-04T21:11:24.530Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 11200, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-complex-work-craftbook-find-vs-create", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-complex-work-craftbook-find-vs-create/craftbook-find-vs-create/craftbook-find-vs-create-underdog-saluki-27b-iq2-mix-2026-11-04T21-37-06-556Z-ntj1", "terminal": true, "logSha256": "65aba89892cb75f4a4aa334ad03ed35f970c67a63ca07fffb9d07b6b5b593e6a", "plans": [ "2026-11-04T21:37:15.443Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 17644, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-complex-work-craftbook-edit-midtask", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-complex-work-craftbook-edit-midtask/craftbook-edit-midtask/craftbook-edit-midtask-underdog-saluki-27b-iq2-mix-2026-11-04T21-41-20-383Z-kl8c", "terminal": true, "logSha256": "ba93a0d6348398d75a4de256543167df188b8b06f94fadccccb27354184fb377", "plans": [ "2026-11-04T21:41:28.156Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 13680, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-complex-work-craftbook-invoice-run", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-complex-work-craftbook-invoice-run/craftbook-invoice-run/craftbook-invoice-run-underdog-saluki-27b-iq2-mix-2026-11-04T23-03-24-956Z-bs2g", "terminal": true, "logSha256": "722ba9efbc5afe43045d6878c95017c251e689f371ee515f142784c8b993ef4a", "plans": [ "2026-11-04T23:03:37.151Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 25936, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-complex-work-craftbook-author-linear", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-complex-work-craftbook-author-linear/craftbook-author-linear/craftbook-author-linear-underdog-saluki-27b-iq2-mix-2026-11-05T00-05-17-288Z-7rlz", "terminal": true, "logSha256": "f8ce50cc296fd02be07293be0c2ee542f06bc7e9bfe90b5bde557a1d4240c570", "plans": [ "2026-11-05T00:05:26.012Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 13592, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-complex-work-craftbook-route-multi", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-complex-work-craftbook-route-multi/craftbook-route-multi/craftbook-route-multi-underdog-saluki-27b-iq2-mix-2026-11-05T00-50-32-680Z-exqp", "terminal": true, "logSha256": "e3efa8f54ca2dd8c6aa925380369b63d7ce20cb08c63d20346267a391607a8ac", "plans": [ "2026-11-05T00:50:39.886Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 43792, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-complex-work-craftbook-export-generalize", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-complex-work-craftbook-export-generalize/craftbook-export-generalize/craftbook-export-generalize-underdog-saluki-27b-iq2-mix-2026-11-05T01-25-52-407Z-2xus", "terminal": true, "logSha256": "146fdfbc8771b54a46348181cb3adaadd66b9f2f06713135e3a73176d04b1cb4", "plans": [ "2026-11-05T01:26:00.640Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 41724, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-complex-work-craftbook-author-params", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-complex-work-craftbook-author-params/craftbook-author-params/craftbook-author-params-underdog-saluki-27b-iq2-mix-2026-11-05T02-38-35-027Z-ok2p", "terminal": true, "logSha256": "8c4185cc1d1b832e31e8f81fe15aeec11827a5389e51218721dfdd75d1f64c3e", "plans": [ "2026-11-05T02:38:43.710Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 50652, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-complex-work-craftbook-author-gate-script", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-complex-work-craftbook-author-gate-script/craftbook-author-gate-script/craftbook-author-gate-script-underdog-saluki-27b-iq2-mix-2026-11-05T03-15-52-588Z-x1tt", "terminal": true, "logSha256": "64cb7256e93505d58d1007a9fe1987d3fcd6f18dbe6ea147d89ee548ad07f6f2", "plans": [ "2026-11-05T03:16:00.869Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 41068, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r1-saluki-cuda-complex-work-craftbook-author-fanout", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r1-saluki-cuda-complex-work-craftbook-author-fanout/craftbook-author-fanout/craftbook-author-fanout-underdog-saluki-27b-iq2-mix-2026-11-05T06-01-02-970Z-sx04", "terminal": true, "logSha256": "efedb7f65dc05ebcff882d2dae329b03a31ce12fe48982d0a086f4ceee157006", "plans": [ "2026-11-05T06:01:12.936Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 25584, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-core-tictactoe", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-core-tictactoe/tictactoe/tictactoe-underdog-saluki-27b-iq2-mix-2026-11-05T09-50-34-101Z-t9wc", "terminal": true, "logSha256": "cbf66a2c5211345bb4967ea7de423faa4829f02f9d197f157001e9747b65c95e", "plans": [ "2026-11-05T09:50:42.569Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 34628, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-core-petshop", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-core-petshop/petshop/petshop-underdog-saluki-27b-iq2-mix-2026-11-05T09-59-21-724Z-fzhl", "terminal": true, "logSha256": "dee4364c3a9470d44c92a88044b171d1b8bb44af8e860803594aa80fe4ba5c05", "plans": [ "2026-11-05T09:59:29.943Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 4244, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" }, { "pid": 48944, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-core-tankcombat", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-core-tankcombat/tankcombat/tankcombat-underdog-saluki-27b-iq2-mix-2026-11-05T11-05-34-374Z-2vk7", "terminal": true, "logSha256": "1acab0ef8daa467bf2658cd6360fe69dbf68f81a861499eb01cc647518085a58", "plans": [ "2026-11-05T11:05:42.360Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 44108, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-core-schema-migration", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-core-schema-migration/schema-migration/schema-migration-underdog-saluki-27b-iq2-mix-2026-11-05T11-15-55-017Z-8hou", "terminal": true, "logSha256": "e5cbfff8418814230ee26e40b810bddf0c70872abb54b15f4c880f505f7449a7", "plans": [ "2026-11-05T11:16:02.772Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 23524, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-core-failing-tests-spec", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-core-failing-tests-spec/failing-tests-spec/failing-tests-spec-underdog-saluki-27b-iq2-mix-2026-11-05T11-27-25-395Z-wslk", "terminal": true, "logSha256": "ee94e34976a6b59db40e97ba18314ec7ac7af8c2c5c41faa1c3726175fe20923", "plans": [ "2026-11-05T11:27:33.937Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 34000, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-core-symptom-debug", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-core-symptom-debug/symptom-debug/symptom-debug-underdog-saluki-27b-iq2-mix-2026-11-05T11-31-24-504Z-k0px", "terminal": true, "logSha256": "2047ec4920a302c6cb079d1540cbb11abf3607ea22e0cbd6eccfd5471cb74d43", "plans": [ "2026-11-05T11:31:32.259Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 25128, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-core-data-wrangle", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-core-data-wrangle/data-wrangle/data-wrangle-underdog-saluki-27b-iq2-mix-2026-11-05T11-37-37-260Z-1m0q", "terminal": true, "logSha256": "be95babd6d725c84f7a479a261311d0013716ab8f14f395972504e0a7f2b8988", "plans": [ "2026-11-05T11:37:44.826Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 37616, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-core-incident-postmortem", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-core-incident-postmortem/incident-postmortem/incident-postmortem-underdog-saluki-27b-iq2-mix-2026-11-05T13-26-14-307Z-y3do", "terminal": true, "logSha256": "5dd2a12c45aa8cad9c441e6af069a42ee5ff81a89d93bdb1a2cf40a0684cbb0c", "plans": [ "2026-11-05T13:26:22.602Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 35396, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-core-ops-runbook-anomaly", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-core-ops-runbook-anomaly/ops-runbook-anomaly/ops-runbook-anomaly-underdog-saluki-27b-iq2-mix-2026-11-05T13-43-51-252Z-xif3", "terminal": true, "logSha256": "08b3f4ad57783eaebb6cb04fe1a7cbaf26de9d4ab39c934291379f6c14df1b36", "plans": [ "2026-11-05T13:43:58.950Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 604, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-core-plan-and-estimate", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-core-plan-and-estimate/plan-and-estimate/plan-and-estimate-underdog-saluki-27b-iq2-mix-2026-11-05T13-52-18-463Z-3yet", "terminal": true, "logSha256": "084a42a18a650d78e828d3b6d25d279411a93dde1d4ef2b5b9f4724b69d42e30", "plans": [ "2026-11-05T13:52:26.862Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 37840, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-core-conflict-synthesis", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-core-conflict-synthesis/conflict-synthesis/conflict-synthesis-underdog-saluki-27b-iq2-mix-2026-11-05T13-58-08-649Z-glwf", "terminal": true, "logSha256": "69dfe6a7d0c3bf1287bb4107143d24e75479935a81eb3ab6e8bc7bc9fdc20763", "plans": [ "2026-11-05T13:58:16.410Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 20316, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" }, { "pid": 2792, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-productivity-constrained-comms", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-productivity-constrained-comms/constrained-comms/constrained-comms-underdog-saluki-27b-iq2-mix-2026-11-05T14-19-48-929Z-ych6", "terminal": true, "logSha256": "9f37b45d34927bbcc4756bf0ba9b08dcd755e667726d47768f5cea9c023e7a5e", "plans": [ "2026-11-05T14:19:57.177Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 34460, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-productivity-records-intake", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-productivity-records-intake/records-intake/records-intake-underdog-saluki-27b-iq2-mix-2026-11-05T14-25-48-976Z-upa4", "terminal": true, "logSha256": "bfcc4d52cec13445148e7fcde506c7649d134f09a3b2c5dd21fdcc5f0913b529", "plans": [ "2026-11-05T14:25:57.199Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 41300, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" }, { "pid": 47268, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-productivity-plan-and-estimate", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-productivity-plan-and-estimate/plan-and-estimate/plan-and-estimate-underdog-saluki-27b-iq2-mix-2026-11-05T16-17-40-854Z-t5k0", "terminal": true, "logSha256": "d9a5a62afdd95542fc2769b1a8b256ebc2fda5cb14e39a72880e9be58af3c3f4", "plans": [ "2026-11-05T16:17:49.503Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 40768, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-productivity-meeting-followup", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-productivity-meeting-followup/meeting-followup/meeting-followup-underdog-saluki-27b-iq2-mix-2026-11-05T16-23-32-328Z-45gu", "terminal": true, "logSha256": "242afd3c41daa00e1646db67ebf1f6fb70948fc72eb57913ca9b697d1799f791", "plans": [ "2026-11-05T16:23:40.950Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 46260, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-productivity-craftbook-spreadsheet-model", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-productivity-craftbook-spreadsheet-model/craftbook-spreadsheet-model/craftbook-spreadsheet-model-underdog-saluki-27b-iq2-mix-2026-11-05T16-31-03-218Z-d6a1", "terminal": true, "logSha256": "97caaf3408e6d9b2a11ec610c5879235d4f978a7d1aaf2eb7f7c6e5161e6820f", "plans": [ "2026-11-05T16:31:14.789Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 36688, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-productivity-conflict-synthesis", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-productivity-conflict-synthesis/conflict-synthesis/conflict-synthesis-underdog-saluki-27b-iq2-mix-2026-11-05T17-48-58-017Z-tjqw", "terminal": true, "logSha256": "b403f3d067589eedc77b911bf7953edfe7807c4eef1172250bfa0d01f8e902ef", "plans": [ "2026-11-05T17:49:06.661Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 38760, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-productivity-docblocks-theme-roundtrip", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-productivity-docblocks-theme-roundtrip/docblocks-theme-roundtrip/docblocks-theme-roundtrip-underdog-saluki-27b-iq2-mix-2026-11-05T17-58-27-608Z-0dux", "terminal": true, "logSha256": "42334ebee64a3a2b7d049f89c0769bf9eaf1869ca43f68ca5376534cae46748b", "plans": [ "2026-11-05T17:58:35.153Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 15472, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-productivity-craftbook-powerpoint-deck", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-productivity-craftbook-powerpoint-deck/craftbook-powerpoint-deck/craftbook-powerpoint-deck-underdog-saluki-27b-iq2-mix-2026-11-05T18-07-11-969Z-2z59", "terminal": true, "logSha256": "56e707f88b935889d77cc455dfdc14eeaa1df725da0e70fecf479c098c53e27d", "plans": [ "2026-11-05T18:07:23.838Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 13092, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-productivity-wikipedia-research-brief", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-productivity-wikipedia-research-brief/wikipedia-research-brief/wikipedia-research-brief-underdog-saluki-27b-iq2-mix-2026-11-05T18-42-31-349Z-vafc", "terminal": true, "logSha256": "1ae28b958d2b747b6801c9f5d0e91e0ca8d9a723f830da0c97068174b3641a49", "plans": [ "2026-11-05T18:42:41.376Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 29676, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-productivity-craftbook-week-plan", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-productivity-craftbook-week-plan/craftbook-week-plan/craftbook-week-plan-underdog-saluki-27b-iq2-mix-2026-11-05T19-02-34-585Z-ivl1", "terminal": true, "logSha256": "7e0345304b8de61633fd6a1350107ce8246364d1722efcc81d4a0c23c7ad9ef2", "plans": [ "2026-11-05T19:02:45.983Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 2320, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-productivity-craftbook-ab-test-readout", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-productivity-craftbook-ab-test-readout/craftbook-ab-test-readout/craftbook-ab-test-readout-underdog-saluki-27b-iq2-mix-2026-11-05T19-37-35-761Z-mx18", "terminal": true, "logSha256": "ef3c4625ffb5a3ce70b974dc5ee1437eba7a6b03637a22fde18c6db93fa01812", "plans": [ "2026-11-05T19:37:47.679Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 35928, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-productivity-craftbook-annotated-bibliography", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-productivity-craftbook-annotated-bibliography/craftbook-annotated-bibliography/craftbook-annotated-bibliography-underdog-saluki-27b-iq2-mix-2026-11-05T21-35-05-918Z-odm0", "terminal": true, "logSha256": "679ac7380e86b806c0dadb5aa42bba8665eddc17296580d32ba76f5f3d3314f2", "plans": [ "2026-11-05T21:35:18.189Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 42472, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-productivity-craftbook-research-to-document", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-productivity-craftbook-research-to-document/craftbook-research-to-document/craftbook-research-to-document-underdog-saluki-27b-iq2-mix-2026-11-05T21-53-08-912Z-e3h0", "terminal": true, "logSha256": "82733005517dab958a631e5061dfada29baa706b3507288fe09cc31b6738b61b", "plans": [ "2026-11-05T21:53:20.565Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 4464, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-developer-fix-squisq-bugs", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-developer-fix-squisq-bugs/fix-squisq-bugs/fix-squisq-bugs-underdog-saluki-27b-iq2-mix-2026-11-05T22-21-46-180Z-mkf3", "terminal": true, "logSha256": "be49d10695adad9357f39b2d7bc9c2d093d8d0ac78dc642e18831cd7041b9cdd", "plans": [ "2026-11-05T22:21:55.028Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 39744, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-developer-dev-craftbook-routing", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-developer-dev-craftbook-routing/dev-craftbook-routing/dev-craftbook-routing-underdog-saluki-27b-iq2-mix-2026-11-05T22-32-46-672Z-wvwm", "terminal": true, "logSha256": "d20c6143f3c51d0e123f710d546413f1802e9fa8271c3745dbf879ebb34aaaf7", "plans": [ "2026-11-05T22:32:54.214Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 39660, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-developer-interface-contract", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-developer-interface-contract/interface-contract/interface-contract-underdog-saluki-27b-iq2-mix-2026-11-05T22-36-51-364Z-ii5y", "terminal": true, "logSha256": "4838fca302271993de3cb7b72990e49996a398985cd744f92e0820b3b0886e5b", "plans": [ "2026-11-05T22:36:58.648Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 44700, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-developer-craftbook-code-review", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-developer-craftbook-code-review/craftbook-code-review/craftbook-code-review-underdog-saluki-27b-iq2-mix-2026-11-05T22-45-20-930Z-ulff", "terminal": true, "logSha256": "028909902cd88b1f6e8f3c9f8dac46ded2e5f69d8ffe7d48869fd2ffb9125c4c", "plans": [ "2026-11-05T22:45:32.868Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 43508, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-developer-craftbook-api-contract-review", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-developer-craftbook-api-contract-review/craftbook-api-contract-review/craftbook-api-contract-review-underdog-saluki-27b-iq2-mix-2026-11-05T23-05-09-444Z-hk2b", "terminal": true, "logSha256": "4c9dc0196484b017d66e5ca916bdaf8ac2e4f654ba3b7071843bb92697801e89", "plans": [ "2026-11-05T23:05:21.183Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 22628, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-developer-craftbook-deep-security-review", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-developer-craftbook-deep-security-review/craftbook-deep-security-review/craftbook-deep-security-review-underdog-saluki-27b-iq2-mix-2026-11-06T00-11-01-045Z-x63v", "terminal": true, "logSha256": "8aa0e26ef954755e3617056160d155cf3207d43d35adce6a38f42cc408878f3d", "plans": [ "2026-11-06T00:11:09.555Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 16572, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-developer-large-pr-review", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-developer-large-pr-review/large-pr-review/large-pr-review-underdog-saluki-27b-iq2-mix-2026-11-06T00-47-11-236Z-gq1s", "terminal": true, "logSha256": "5000ce9c756e00f1dce96c6f6eb71b8c67f0c4444f18c6bc7079e9e3f8688818", "plans": [ "2026-11-06T00:47:20.567Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 35896, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-developer-craftbook-codemod-sweep", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-developer-craftbook-codemod-sweep/craftbook-codemod-sweep/craftbook-codemod-sweep-underdog-saluki-27b-iq2-mix-2026-11-06T02-32-28-346Z-cu6z", "terminal": true, "logSha256": "237ee8e885566277064536504a86e85246801b5a539a9409ebe843003b4b8624", "plans": [ "2026-11-06T02:32:41.748Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 36660, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-developer-craftbook-refactor-module", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-developer-craftbook-refactor-module/craftbook-refactor-module/craftbook-refactor-module-underdog-saluki-27b-iq2-mix-2026-11-06T03-07-31-670Z-u16c", "terminal": true, "logSha256": "00aaf6c84d03f0440d465011fdc290a6e2e10f6a8bc096e3e771599252936c51", "plans": [ "2026-11-06T03:07:44.076Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 33552, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-developer-craftbook-bug-fix-tdd", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-developer-craftbook-bug-fix-tdd/craftbook-bug-fix-tdd/craftbook-bug-fix-tdd-underdog-saluki-27b-iq2-mix-2026-11-06T04-12-52-338Z-lfv4", "terminal": true, "logSha256": "22c4ad5a1f974226c0046e52720182659b6272e1276e20e25089aa989565c0e8", "plans": [ "2026-11-06T04:13:05.741Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 36536, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-complex-work-craftbook-find-vs-create", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-complex-work-craftbook-find-vs-create/craftbook-find-vs-create/craftbook-find-vs-create-underdog-saluki-27b-iq2-mix-2026-11-06T04-40-51-154Z-xkhe", "terminal": true, "logSha256": "9fa8bbb371ccec58920003138a9fc1a66b5a7a1b0c59537cd9e4e116e3a68aca", "plans": [ "2026-11-06T04:40:59.511Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 2900, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-complex-work-craftbook-edit-midtask", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-complex-work-craftbook-edit-midtask/craftbook-edit-midtask/craftbook-edit-midtask-underdog-saluki-27b-iq2-mix-2026-11-06T04-45-07-147Z-ymas", "terminal": true, "logSha256": "7777dd53da733c66ebc274d298e99084caff923e535d020321c307bf0613b225", "plans": [ "2026-11-06T04:45:14.869Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 42416, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-complex-work-craftbook-invoice-run", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-complex-work-craftbook-invoice-run/craftbook-invoice-run/craftbook-invoice-run-underdog-saluki-27b-iq2-mix-2026-11-06T05-51-45-280Z-gkvw", "terminal": true, "logSha256": "065c4de582147a464a3c0518b0a89a862e1ae79940b26ef1cf36bf7685a0c744", "plans": [ "2026-11-06T05:51:57.471Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 53004, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-complex-work-craftbook-author-linear", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-complex-work-craftbook-author-linear/craftbook-author-linear/craftbook-author-linear-underdog-saluki-27b-iq2-mix-2026-11-06T06-22-58-627Z-pazz", "terminal": true, "logSha256": "73628735500926ef349dbf2b1d894421e4f6e282e75d738e142c5fb2543f6bd7", "plans": [ "2026-11-06T06:23:07.183Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 47876, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-complex-work-craftbook-route-multi", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-complex-work-craftbook-route-multi/craftbook-route-multi/craftbook-route-multi-underdog-saluki-27b-iq2-mix-2026-11-06T07-33-08-377Z-788t", "terminal": true, "logSha256": "cb0edb92d515d24df5b09f2afd242d9f0903f4d74f01ea1320210f3b25ca88f5", "plans": [ "2026-11-06T07:33:18.220Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 41416, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-complex-work-craftbook-export-generalize", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-complex-work-craftbook-export-generalize/craftbook-export-generalize/craftbook-export-generalize-underdog-saluki-27b-iq2-mix-2026-11-06T08-10-28-851Z-anv3", "terminal": true, "logSha256": "1e04d6c4296236c4335fcdef4343984484ee064736b3d4e0e4e5962c2e86a55d", "plans": [ "2026-11-06T08:10:37.677Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 48220, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-complex-work-craftbook-author-params", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-complex-work-craftbook-author-params/craftbook-author-params/craftbook-author-params-underdog-saluki-27b-iq2-mix-2026-11-06T14-54-26-198Z-9jmc", "terminal": true, "logSha256": "b711d6071e3655197caa814b32128b5f68b995844f4cfb77d68c31ca204e2410", "plans": [ "2026-11-06T14:54:35.193Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 7856, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-complex-work-craftbook-author-gate-script", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/r2-saluki-cuda-complex-work-craftbook-author-gate-script/craftbook-author-gate-script/craftbook-author-gate-script-underdog-saluki-27b-iq2-mix-2026-11-06T15-16-17-977Z-jsdg", "terminal": true, "logSha256": "ae0782b5f1d4180744c5044ce65b9aee8d258265f86a8c1d160c259defe20b20", "plans": [ "2026-11-06T15:16:26.778Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 38712, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" }, { "pid": 39856, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-core-tictactoe", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-core-tictactoe/tictactoe/tictactoe-underdog-saluki-27b-iq2-mix-2026-11-09T20-16-26-587Z-xifl", "terminal": true, "logSha256": "ff2a4b0f9a8212b708eaa2af5874834648c8837b6f8293dd84aaec87e3f43616", "plans": [ "2026-11-09T20:16:35.320Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 26008, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-core-petshop", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-core-petshop/petshop/petshop-underdog-saluki-27b-iq2-mix-2026-11-09T20-25-10-813Z-b8dy", "terminal": true, "logSha256": "3a713ec8e50963a1587c4f14eae97daa22db40e8570d5906ae6c96376af58e25", "plans": [ "2026-11-09T20:25:18.242Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 42660, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" }, { "pid": 23492, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-core-tankcombat", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-core-tankcombat/tankcombat/tankcombat-underdog-saluki-27b-iq2-mix-2026-11-09T21-14-42-224Z-41w8", "terminal": true, "logSha256": "c9e9a047e62482866e84e8b8ef67b510e07d9e51b2b2cd73d790aaed0be3546a", "plans": [ "2026-11-09T21:14:49.671Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 36352, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-core-schema-migration", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-core-schema-migration/schema-migration/schema-migration-underdog-saluki-27b-iq2-mix-2026-11-09T21-25-46-082Z-a203", "terminal": true, "logSha256": "34817bcefbf6c6f789ceb5cafc8b42c38735f7c2108b8e8a2fc3e9cfe70a8c67", "plans": [ "2026-11-09T21:25:53.927Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 21440, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-core-failing-tests-spec", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-core-failing-tests-spec/failing-tests-spec/failing-tests-spec-underdog-saluki-27b-iq2-mix-2026-11-09T21-37-05-190Z-4vve", "terminal": true, "logSha256": "6983c06dec6e1b1afc12afe59e578ff188bbf43b425ae287075659281c995f80", "plans": [ "2026-11-09T21:37:13.127Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 34920, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-core-symptom-debug", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-core-symptom-debug/symptom-debug/symptom-debug-underdog-saluki-27b-iq2-mix-2026-11-09T21-41-18-850Z-2jfn", "terminal": true, "logSha256": "25662b0109144b99420581fcb08ac3f6b2977a6521e530459e4bfdd276c4d412", "plans": [ "2026-11-09T21:41:26.459Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 24780, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-core-data-wrangle", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-core-data-wrangle/data-wrangle/data-wrangle-underdog-saluki-27b-iq2-mix-2026-11-09T21-51-35-159Z-obd0", "terminal": true, "logSha256": "6894720cd76f8e517c589b9767a4fb108bd3c57b314077cf188cedd7a4f6936b", "plans": [ "2026-11-09T21:51:42.954Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 54112, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-core-incident-postmortem", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-core-incident-postmortem/incident-postmortem/incident-postmortem-underdog-saluki-27b-iq2-mix-2026-11-09T21-58-34-962Z-p5hj", "terminal": true, "logSha256": "ba511115c95c9298bc4d73030bb447fe85d37f4cd6c6d4c24dca46c37bee0753", "plans": [ "2026-11-09T21:58:42.361Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 49088, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-core-ops-runbook-anomaly", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-core-ops-runbook-anomaly/ops-runbook-anomaly/ops-runbook-anomaly-underdog-saluki-27b-iq2-mix-2026-11-09T22-16-29-575Z-tvu2", "terminal": true, "logSha256": "18285640b75afdc01144764fbb76b46da05cb0a66e584b945e8c0013d8cff051", "plans": [ "2026-11-09T22:16:37.133Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 42584, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-core-plan-and-estimate", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-core-plan-and-estimate/plan-and-estimate/plan-and-estimate-underdog-saluki-27b-iq2-mix-2026-11-09T22-23-01-777Z-g3kt", "terminal": true, "logSha256": "e9a37129fa0d975918e5034c8c612a352388c37501de5f9273b42f405cfc0787", "plans": [ "2026-11-09T22:23:09.510Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 36068, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-core-conflict-synthesis", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-core-conflict-synthesis/conflict-synthesis/conflict-synthesis-underdog-saluki-27b-iq2-mix-2026-11-09T22-28-21-713Z-3kl0", "terminal": true, "logSha256": "864bedec2e24eb178801c3a5e4a6f00eb24199a98bf30f353300a72878042677", "plans": [ "2026-11-09T22:28:29.290Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 42780, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" }, { "pid": 14020, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-productivity-constrained-comms", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-productivity-constrained-comms/constrained-comms/constrained-comms-underdog-saluki-27b-iq2-mix-2026-11-09T22-59-25-389Z-yth7", "terminal": true, "logSha256": "a77305d8edf32115cbdb8c4aad1898fe26ae21296d06b64286a6db62b6c10ffb", "plans": [ "2026-11-09T22:59:33.218Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 12844, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-productivity-records-intake", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-productivity-records-intake/records-intake/records-intake-underdog-saluki-27b-iq2-mix-2026-11-09T23-04-41-731Z-tlqe", "terminal": true, "logSha256": "93cb3777550394534b39340356285fc9d5a2048132a428da0687d8be00b9c2ce", "plans": [ "2026-11-09T23:04:49.177Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 18400, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-productivity-plan-and-estimate", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-productivity-plan-and-estimate/plan-and-estimate/plan-and-estimate-underdog-saluki-27b-iq2-mix-2026-11-10T00-23-29-233Z-ul8o", "terminal": true, "logSha256": "a216f4d17d3708f39528e03c47b7669056aa56f01a8033312c32f68a4b313fbf", "plans": [ "2026-11-10T00:23:36.818Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 40956, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-productivity-meeting-followup", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-productivity-meeting-followup/meeting-followup/meeting-followup-underdog-saluki-27b-iq2-mix-2026-11-10T00-29-08-898Z-1nl3", "terminal": true, "logSha256": "131bb44348f9d1c4bc31217e9e5f1f9f62428ccccb43d0dd52ff2e546a227e02", "plans": [ "2026-11-10T00:29:16.916Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 15032, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-productivity-craftbook-spreadsheet-model", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-productivity-craftbook-spreadsheet-model/craftbook-spreadsheet-model/craftbook-spreadsheet-model-underdog-saluki-27b-iq2-mix-2026-11-10T00-36-23-881Z-33tc", "terminal": true, "logSha256": "a12b613448f2ff4a906a70ab1544448af8b529899e8410c619f76ad27af384ac", "plans": [ "2026-11-10T00:36:35.381Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 4464, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-productivity-conflict-synthesis", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-productivity-conflict-synthesis/conflict-synthesis/conflict-synthesis-underdog-saluki-27b-iq2-mix-2026-11-10T02-09-59-228Z-9zyu", "terminal": true, "logSha256": "15dcb0245aa4bbd8fd34352d25b77b298d6fed78bcaa62a07696929040307e8d", "plans": [ "2026-11-10T02:10:09.140Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 23332, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-productivity-docblocks-theme-roundtrip", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-productivity-docblocks-theme-roundtrip/docblocks-theme-roundtrip/docblocks-theme-roundtrip-underdog-saluki-27b-iq2-mix-2026-11-10T02-43-21-395Z-w8ls", "terminal": true, "logSha256": "3e786bda96ca31f726889c756903526b9e0b77437dffeb0e3cd48bfabf40e1ce", "plans": [ "2026-11-10T02:43:31.214Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 39960, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" }, { "pid": 47088, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-productivity-craftbook-powerpoint-deck", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-productivity-craftbook-powerpoint-deck/craftbook-powerpoint-deck/craftbook-powerpoint-deck-underdog-saluki-27b-iq2-mix-2026-11-10T03-05-12-731Z-u76b", "terminal": true, "logSha256": "05847f409418823445570e838ed5ef6eac64e02920e6a0bf2629ea83783b4820", "plans": [ "2026-11-10T03:05:25.941Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 33880, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-productivity-wikipedia-research-brief", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-productivity-wikipedia-research-brief/wikipedia-research-brief/wikipedia-research-brief-underdog-saluki-27b-iq2-mix-2026-11-10T04-37-26-781Z-v7zl", "terminal": true, "logSha256": "cec3d98e5ae0a7c427ccbd6e4fc6b9f9bff48a01e24ee69947b0c78919c4c8bd", "plans": [ "2026-11-10T04:37:39.417Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 33568, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-productivity-craftbook-week-plan", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-productivity-craftbook-week-plan/craftbook-week-plan/craftbook-week-plan-underdog-saluki-27b-iq2-mix-2026-11-10T05-31-35-469Z-qnfz", "terminal": true, "logSha256": "fee1af3ad1ab37848589100171f04fcc8b6eb0d958fdc5aa5c016a4ed854781b", "plans": [ "2026-11-10T05:31:48.724Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 47940, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-productivity-craftbook-ab-test-readout", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-productivity-craftbook-ab-test-readout/craftbook-ab-test-readout/craftbook-ab-test-readout-underdog-saluki-27b-iq2-mix-2026-11-10T06-57-26-407Z-3wb1", "terminal": true, "logSha256": "8e3658f890ff56055517c64de690e57ff96585f34bacbf136ca57249d8eba4f9", "plans": [ "2026-11-10T06:57:40.899Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 21528, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-productivity-craftbook-annotated-bibliography", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-productivity-craftbook-annotated-bibliography/craftbook-annotated-bibliography/craftbook-annotated-bibliography-underdog-saluki-27b-iq2-mix-2026-11-10T08-08-39-484Z-f8cd", "terminal": true, "logSha256": "a2647da17b3c89f47e5e3dc6f811638e3afaadce8226a2ef79575a45ffdad54f", "plans": [ "2026-11-10T08:08:52.539Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 53576, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-productivity-craftbook-research-to-document", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-productivity-craftbook-research-to-document/craftbook-research-to-document/craftbook-research-to-document-underdog-saluki-27b-iq2-mix-2026-11-10T08-53-56-539Z-295e", "terminal": true, "logSha256": "5850e9639376af6ee2d13db6527b28f9e54237e4bc2305e63acba8d704a1b41a", "plans": [ "2026-11-10T08:54:09.844Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 28076, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-developer-fix-squisq-bugs", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-developer-fix-squisq-bugs/fix-squisq-bugs/fix-squisq-bugs-underdog-saluki-27b-iq2-mix-2026-11-10T09-40-09-307Z-no9t", "terminal": true, "logSha256": "9d24d28b57492dc35d98ad8be5147bd6d4bb2d8ff83384a174a878a8f1ed27e0", "plans": [ "2026-11-10T09:40:18.996Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 37092, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-developer-dev-craftbook-routing", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-developer-dev-craftbook-routing/dev-craftbook-routing/dev-craftbook-routing-underdog-saluki-27b-iq2-mix-2026-11-10T10-06-52-675Z-5gvb", "terminal": true, "logSha256": "1c4a3bbcb7740274f9bf258efc71a9181f424568af626eb41f63fe5c65c5f36a", "plans": [ "2026-11-10T10:07:02.245Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 31092, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-developer-interface-contract", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-developer-interface-contract/interface-contract/interface-contract-underdog-saluki-27b-iq2-mix-2026-11-10T10-13-02-629Z-xihj", "terminal": true, "logSha256": "51f710e57894118e94c9ada7fc2b8c9de4c41f929fef850b24e2cdbef9b3a159", "plans": [ "2026-11-10T10:13:11.637Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 9852, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-developer-craftbook-code-review", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-developer-craftbook-code-review/craftbook-code-review/craftbook-code-review-underdog-saluki-27b-iq2-mix-2026-11-10T10-29-58-625Z-kelb", "terminal": true, "logSha256": "083874f6bda2dd89518401a52ea6cead03f374bcc2a40c87220ff9ba968a14b6", "plans": [ "2026-11-10T10:30:11.668Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 27076, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-developer-craftbook-api-contract-review", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-developer-craftbook-api-contract-review/craftbook-api-contract-review/craftbook-api-contract-review-underdog-saluki-27b-iq2-mix-2026-11-10T11-00-03-208Z-gfja", "terminal": true, "logSha256": "3ea7e6565de1d6c23af28841af75f93681f30a4dfe864cb7cba641c1fc131ba6", "plans": [ "2026-11-10T11:00:16.264Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 12924, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-developer-craftbook-deep-security-review", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-developer-craftbook-deep-security-review/craftbook-deep-security-review/craftbook-deep-security-review-underdog-saluki-27b-iq2-mix-2026-11-10T11-56-39-229Z-cava", "terminal": true, "logSha256": "a8000535ddb590d89b00108af375e837e94cfe8b7956f3166e5daa3b879d8523", "plans": [ "2026-11-10T11:56:48.817Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 44668, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-developer-large-pr-review", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-developer-large-pr-review/large-pr-review/large-pr-review-underdog-saluki-27b-iq2-mix-2026-11-10T12-50-20-834Z-9d29", "terminal": true, "logSha256": "ce3ecbc3f2329a7e4851979c864b2c9969b90c324f63b0ce3bb6989994b0ffa9", "plans": [ "2026-11-10T12:50:31.656Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 40876, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-developer-craftbook-codemod-sweep", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-developer-craftbook-codemod-sweep/craftbook-codemod-sweep/craftbook-codemod-sweep-underdog-saluki-27b-iq2-mix-2026-11-10T13-17-14-802Z-tdv3", "terminal": true, "logSha256": "1eb9574cbeb8a112d14acd4914c283103c46857c1ff1ee7872f3c1a3ab3a6efa", "plans": [ "2026-11-10T13:17:27.885Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 44312, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-developer-craftbook-refactor-module", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-developer-craftbook-refactor-module/craftbook-refactor-module/craftbook-refactor-module-underdog-saluki-27b-iq2-mix-2026-11-10T14-28-09-481Z-ilo4", "terminal": true, "logSha256": "356d89615f012894963d5f00c9cea4580a7e8dcd0eee4e1ba3201140b46a065f", "plans": [ "2026-11-10T14:28:22.537Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 30056, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-developer-craftbook-bug-fix-tdd", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-developer-craftbook-bug-fix-tdd/craftbook-bug-fix-tdd/craftbook-bug-fix-tdd-underdog-saluki-27b-iq2-mix-2026-11-10T15-19-58-344Z-ilty", "terminal": true, "logSha256": "2099930699c9b7dccbf3ac4b7d7afc393ec0d9cbee9cd0fcd19523c3df82826c", "plans": [ "2026-11-10T15:20:11.433Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 40520, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-complex-work-craftbook-find-vs-create", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-complex-work-craftbook-find-vs-create/craftbook-find-vs-create/craftbook-find-vs-create-underdog-saluki-27b-iq2-mix-2026-11-10T16-22-59-365Z-ikxq", "terminal": true, "logSha256": "db5511124368eb487e458e2b402bbc0147e9b16dfede5a26c16d4605ac853f2c", "plans": [ "2026-11-10T16:23:08.896Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 17756, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-complex-work-craftbook-edit-midtask", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-complex-work-craftbook-edit-midtask/craftbook-edit-midtask/craftbook-edit-midtask-underdog-saluki-27b-iq2-mix-2026-11-10T16-28-41-233Z-pq67", "terminal": true, "logSha256": "170f995578c0c2ba86f4e152eb1acdff0f1c28f19b270741e9b8645e44df59c7", "plans": [ "2026-11-10T16:28:50.388Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 38056, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-complex-work-craftbook-invoice-run", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-complex-work-craftbook-invoice-run/craftbook-invoice-run/craftbook-invoice-run-underdog-saluki-27b-iq2-mix-2026-11-10T16-55-17-811Z-d5ml", "terminal": true, "logSha256": "6ac18709b905d76be67fd3f101c10ae72407602544d2bd4bd09e0d615741179c", "plans": [ "2026-11-10T16:55:29.788Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 11508, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-complex-work-craftbook-author-linear", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-complex-work-craftbook-author-linear/craftbook-author-linear/craftbook-author-linear-underdog-saluki-27b-iq2-mix-2026-11-10T17-52-16-268Z-99gm", "terminal": true, "logSha256": "27a97a2df81871e6c14b48f9c203fc4f0d34dbc5277dc4d6690afd5062194fbb", "plans": [ "2026-11-10T17:52:24.652Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 25096, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-complex-work-craftbook-route-multi", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-complex-work-craftbook-route-multi/craftbook-route-multi/craftbook-route-multi-underdog-saluki-27b-iq2-mix-2026-11-10T18-52-14-365Z-dhdr", "terminal": true, "logSha256": "d6a24b70af5b002210df05159c71ecd10492b844684e107c358eb0778f3b2ce6", "plans": [ "2026-11-10T18:52:22.876Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 47896, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-complex-work-craftbook-export-generalize", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-complex-work-craftbook-export-generalize/craftbook-export-generalize/craftbook-export-generalize-underdog-saluki-27b-iq2-mix-2026-11-10T19-54-06-807Z-50yi", "terminal": true, "logSha256": "10cc46d79fb28a8c37d479a9f4ee477b66818920d9e88b3639c36b64c137a388", "plans": [ "2026-11-10T19:54:14.808Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 34488, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-complex-work-craftbook-author-params", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-complex-work-craftbook-author-params/craftbook-author-params/craftbook-author-params-underdog-saluki-27b-iq2-mix-2026-11-10T22-28-51-912Z-754f", "terminal": true, "logSha256": "e31a83c465f0f53392fcf60edf57117079d87b28c7b21ba5dfe52711c75d7b17", "plans": [ "2026-11-10T22:28:59.456Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 44808, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r2-saluki-cuda-complex-work-craftbook-author-fanout", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/fanout-remainder/craftbook-author-fanout/craftbook-author-fanout-underdog-saluki-27b-iq2-mix-2026-11-06T19-54-43-767Z-b8ob", "terminal": false, "logSha256": "9d6d340d90e2162e266a99e464cfc67b51b1bb62da9386d6ae0233c1d29e7cff", "plans": [ "2026-11-06T19:54:51.464Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 33840, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] }, { "cellId": "r3-saluki-cuda-complex-work-craftbook-author-gate-script", "trialDir": "evals/runs/saluki-native-0.1.57-2026-11-03/continuation-2026-11-06/round3/r1-saluki-cuda-complex-work-craftbook-author-gate-script/craftbook-author-gate-script/craftbook-author-gate-script-underdog-saluki-27b-iq2-mix-2026-11-11T00-06-24-814Z-1294", "terminal": false, "logSha256": "8cbd56f2a12edd06c989c549680a3c589b2249c2cd23d3a398b3bb745528de15", "plans": [ "2026-11-11T00:06:33.235Z INFO [chat] [llama-cpp] offload plan (underdog-saluki-27b-iq2-mix): dense FFN of 39/64 layers fits VRAM (3.0 GiB of 4.5 GiB, non-FFN 2.9 GiB, reserves 5.5 GiB) — --n-cpu-ffn 25" ], "launches": [ { "pid": 49480, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" }, { "pid": 51964, "command": "gezel-llama-server.exe", "nativeRelease": "0.1.57", "model": "underdog-saluki-27b-iq2-mix", "backend": "cuda", "upstreamRevision": "8642541b255bfb5dc39cd351a178be09013a55d6", "cudaArchitectures": "60-virtual;70-real;75-real;80-real;86-real;89-real;90-real;100-real;120", "cudaToolkit": "12.8", "gpu": "NVIDIA GeForce RTX 5070 Ti Laptop GPU", "computeCapability": "12.0", "driverVersion": "610.88", "contextPerSlot": 65536, "contextTotal": 65536, "slots": 1, "kvCacheType": "f16", "batchSize": "default", "ubatchSize": "512", "flashAttention": "default", "threads": "default", "reasoningBudgetTokens": 4096, "reasoningBudgetSource": "catalog", "reasoningPreserve": false, "cudaGraphs": "default" } ], "gpuCpuPairs": [ { "gpu": 39, "total": 64, "cpu": 25 } ] } ] }