{ "schema": 1, "parameters": { "werk": "target/debug/werk", "config": "/tmp/werk-mgpu-single.json", "out": "/tmp/werk-mgpu-single-results.json", "repetitions": 3, "tokens": 16, "prompt": "Return the integer result of 17 + 25, with no explanation.", "required_gpus": 1, "timeout": 600 }, "inventory_before": { "devices": [ { "id": "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9", "physical_index": 0, "vendor": "nvidia", "name": "NVIDIA GeForce RTX 3090", "architecture": "sm_86", "runtime": "cuda", "pci_bus_id": "00000000:03:00.0", "total_bytes": 25769803776, "free_bytes": 21704474624, "numa_node": null, "pcie_link": null, "visible_index": 0 } ], "host_available_bytes": 96888598528, "topology": null, "warnings": [] }, "werk_revision": "8d8ee73a002e8473818d1e86099e0a21b3b51d23", "dirty_diff": " .gitignore | 3 +-\n src/api.rs | 1 +\n src/api/chat.rs | 44 ++++++++++--\n src/api/extended.rs | 22 ++++--\n src/api/generation.rs | 26 ++++++-\n src/api/observability.rs | 3 +-\n src/api/router.rs | 1 +\n src/api/state.rs | 48 +++++++++++++\n src/api/werk/mod.rs | 40 +++++++++++\n src/backend/llama_process_lifecycle.rs | 34 +++++++--\n src/backend/llama_server.rs | 77 ++++++++++++++++++--\n src/backend/llama_server/fp4.rs | 8 +++\n src/backend/llama_server/runtime_state.rs | 116 ++++++++++++++++--------------\n src/backend/vllm.rs | 76 +++++++++++++++++++-\n src/backend/vllm/runtime_control.rs | 6 +-\n src/cli.rs | 59 ++++++++++++++-\n src/inference_service.rs | 3 +-\n src/inference_service/resources.rs | 92 ++++++++++++++++++++++++\n src/lib.rs | 1 +\n 19 files changed, 574 insertions(+), 86 deletions(-)\n", "cold_definition": "fresh worker; OS page cache not flushed", "samples": [ { "alias": "qwen-gpu", "session": "validation-qwen-gpu", "ttft_seconds": 28.105974994999997, "total_seconds": 28.28231289299947, "usage": { "completion_tokens": 3, "prompt_tokens": 29, "prompt_tokens_details": { "cached_tokens": 0 }, "total_tokens": 32 }, "text": "42", "repetition": 0, "worker_state": "cold" }, { "alias": "qwen-gpu", "session": "validation-qwen-gpu", "ttft_seconds": 0.8999330920014472, "total_seconds": 1.022885390000738, "usage": { "completion_tokens": 3, "prompt_tokens": 29, "prompt_tokens_details": { "cached_tokens": 25 }, "total_tokens": 32 }, "text": "42", "repetition": 1, "worker_state": "warm" }, { "alias": "qwen-gpu", "session": "validation-qwen-gpu", "ttft_seconds": 0.5221091950006667, "total_seconds": 0.609684594997816, "usage": { "completion_tokens": 3, "prompt_tokens": 29, "prompt_tokens_details": { "cached_tokens": 25 }, "total_tokens": 32 }, "text": "42", "repetition": 2, "worker_state": "warm" } ], "hardware_cases": { "single": "configured", "two_equal": "requires explicit config and >=2 GPUs", "two_unequal": "requires explicit config and >=2 unequal GPUs", "independent": "requires explicit config and >=2 GPUs", "mixed": "requires explicit config and >=3 GPUs", "replicas": "requires explicit config and >=2 GPUs" }, "plans": [ { "profile": { "id": "qwen-gpu", "alias": "multi-gpu-smoke", "model": "qwen3.8-27b-ud-q4_k_xl", "runtime": "llama_cuda", "executable": "/home/philipbodenbach/.local/share/werk1112/backends/llama-cuda/upstream-ec928150501c/build/bin/llama-server", "build": "ec928150501c", "gpus": [ "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9" ], "strategy": "single", "tensor_parallel": 1, "pipeline_parallel": 1, "split": [], "context": 2048, "batch": 256, "parallel": 1, "cpu_threads": 8, "offload": { "kind": "none" }, "kv_cpu": false, "memory": [ { "weights": 18500000000, "replicated": 0, "kv": 268435456, "experts": 0, "compute": 536870912, "overhead": 268435456, "reserve": 536870912 } ], "host_bytes": 24000000000, "native_args": [ "--no-warmup" ] }, "model_identity": "sha256:1352bb293f26ccf8ee2fd8d226093eff8357e5eac21f078dca9f0e8bc2f6f071", "executable_sha256": "236d00dc1c71b01bd15a13b229edeac2e4b05a9ac740d33aadc14fe6f71287b6", "devices": [ { "id": "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9", "physical_index": 0, "vendor": "nvidia", "name": "NVIDIA GeForce RTX 3090", "architecture": "sm_86", "runtime": "cuda", "pci_bus_id": "00000000:03:00.0", "total_bytes": 25769803776, "free_bytes": 21721251840, "numa_node": null, "pcie_link": null, "visible_index": 0 } ], "strategy": "single", "args": [ "--device", "CUDA0", "--split-mode", "none", "--main-gpu", "0", "--no-warmup" ], "environment": { "CUDA_DEVICE_ORDER": "PCI_BUS_ID", "CUDA_VISIBLE_DEVICES": "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9", "OMP_NUM_THREADS": "8" }, "fingerprint": "5878014b255b22aa490568d6cf78ad3db30a2c28be70b60414154c5fac06090a", "reasons": [ "Exclusive GPU ownership; component budgets are declared estimates, not measured usage or hard allocator limits.", "Unknown P2P/NVLink is not treated as a prerequisite; no throughput prediction was made." ] } ], "telemetry": [ { "protocol": { "major": 1, "minor": 0 }, "request_id": "req_N3AIu3d25fQByRw6V5ws_A", "data": { "host_swap_used_bytes": 0, "host_memory_free_bytes": 73435750400, "schema_version": 1, "observed_at_ms": 1791310703333, "server_started_ms": 1791310672545, "uptime_seconds": 30.78783743, "totals": { "started": 1, "completed": 1, "errors": 0, "cancelled": 0, "active": 0, "prompt_tokens": 29, "output_tokens": 3, "cached_tokens": 0, "duration_seconds": 27.999049124 }, "requests": [ { "id": 1, "model": "qwen3.8-27b-ud-q4_k_xl", "state": "done", "started_ms": 1791310675215, "elapsed_seconds": 27.999049124, "prompt_tokens": 29, "cached_tokens": 0, "output_tokens": 3, "first_output_seconds": 0.432180295, "decode_tokens_per_second": 16.981676770764345, "prefill_tokens_per_second": 67.7033844688436 } ], "backends": [ { "backend": "deployment", "instance": "qwen-gpu", "model": "multi-gpu-smoke", "available": true, "counters": {}, "gauges": { "gpu_budget_bytes_estimate": 20110612736.0, "gpu_count": 1.0, "host_budget_bytes_estimate": 24000000000.0, "queue_capacity": 0.0, "slots_available": 1.0 } }, { "backend": "llama.cpp / CUDA", "instance": "llama-55319-http://127.0.0.1:43411", "model": "qwen3.8-27b-ud-q4_k_xl", "available": true, "counters": {}, "gauges": { "active_cached_tokens": 0.0, "active_output_tokens": 0.0, "active_prompt_tokens": 0.0, "context_capacity_tokens": 2048.0, "context_used_tokens": 31.0, "gpu_layers_requested": 999.0, "process_resident_bytes": 1456742400.0, "requests_active": 0.0, "slots_total": 1.0 } } ], "memory": { "observed_at_unix_ms": 1791310703332, "overall_pressure": "hard", "topology": "discrete", "host": { "capacity_bytes": 105469059072, "available_bytes": 99107876864, "managed_bytes": 0, "reserved_bytes": 0, "pressure": "normal" }, "accelerator": { "capacity_bytes": 25769803776, "available_bytes": 3666870272, "managed_bytes": 0, "reserved_bytes": 0, "pressure": "hard" }, "last_action_unix_ms": null, "counters": { "active_reservations": 0, "backend_cleanup_failures": 0, "backend_cleanup_latched": 0, "completed_demotions": 0, "completed_evictions": 0, "failed_releases": 0, "managed_allocations": 0, "orphaned_release_bytes": 0, "pressure_actions_in_flight": 0 } }, "backend_observed_at_ms": 1791310703226 } }, { "protocol": { "major": 1, "minor": 0 }, "request_id": "req_YrEnM1zfFIBa5ZmYF6yRcA", "data": { "host_swap_used_bytes": 0, "host_memory_free_bytes": 73435750400, "schema_version": 1, "observed_at_ms": 1791310704360, "server_started_ms": 1791310672545, "uptime_seconds": 31.81538422, "totals": { "started": 2, "completed": 2, "errors": 0, "cancelled": 0, "active": 0, "prompt_tokens": 58, "output_tokens": 6, "cached_tokens": 25, "duration_seconds": 28.721183016999998 }, "requests": [ { "id": 2, "model": "qwen3.8-27b-ud-q4_k_xl", "state": "done", "started_ms": 1791310703634, "elapsed_seconds": 0.722133893, "prompt_tokens": 29, "cached_tokens": 25, "output_tokens": 3, "first_output_seconds": 0.597317795, "decode_tokens_per_second": 24.646527715020415, "prefill_tokens_per_second": 14.992447554544398 }, { "id": 1, "model": "qwen3.8-27b-ud-q4_k_xl", "state": "done", "started_ms": 1791310675215, "elapsed_seconds": 27.999049124, "prompt_tokens": 29, "cached_tokens": 0, "output_tokens": 3, "first_output_seconds": 0.432180295, "decode_tokens_per_second": 16.981676770764345, "prefill_tokens_per_second": 67.7033844688436 } ], "backends": [ { "backend": "deployment", "instance": "qwen-gpu", "model": "multi-gpu-smoke", "available": true, "counters": {}, "gauges": { "gpu_budget_bytes_estimate": 20110612736.0, "gpu_count": 1.0, "host_budget_bytes_estimate": 24000000000.0, "queue_capacity": 0.0, "slots_available": 1.0 } }, { "backend": "llama.cpp / CUDA", "instance": "llama-55319-http://127.0.0.1:43411", "model": "qwen3.8-27b-ud-q4_k_xl", "available": true, "counters": {}, "gauges": { "active_cached_tokens": 0.0, "active_output_tokens": 0.0, "active_prompt_tokens": 0.0, "context_capacity_tokens": 2048.0, "context_used_tokens": 31.0, "gpu_layers_requested": 999.0, "process_resident_bytes": 1456742400.0, "requests_active": 0.0, "slots_total": 1.0 } } ], "memory": { "observed_at_unix_ms": 1791310703332, "overall_pressure": "hard", "topology": "discrete", "host": { "capacity_bytes": 105469059072, "available_bytes": 99107876864, "managed_bytes": 0, "reserved_bytes": 0, "pressure": "normal" }, "accelerator": { "capacity_bytes": 25769803776, "available_bytes": 3666870272, "managed_bytes": 0, "reserved_bytes": 0, "pressure": "hard" }, "last_action_unix_ms": null, "counters": { "active_reservations": 0, "backend_cleanup_failures": 0, "backend_cleanup_latched": 0, "completed_demotions": 0, "completed_evictions": 0, "failed_releases": 0, "managed_allocations": 0, "orphaned_release_bytes": 0, "pressure_actions_in_flight": 0 } }, "backend_observed_at_ms": 1791310703226 } }, { "protocol": { "major": 1, "minor": 0 }, "request_id": "req_gKKEOPaOr_xOFM0eWMRN7Q", "data": { "host_swap_used_bytes": 0, "host_memory_free_bytes": 73435750400, "schema_version": 1, "observed_at_ms": 1791310704974, "server_started_ms": 1791310672545, "uptime_seconds": 32.429616115, "totals": { "started": 3, "completed": 3, "errors": 0, "cancelled": 0, "active": 0, "prompt_tokens": 87, "output_tokens": 9, "cached_tokens": 50, "duration_seconds": 29.058667313999997 }, "requests": [ { "id": 3, "model": "qwen3.8-27b-ud-q4_k_xl", "state": "done", "started_ms": 1791310704634, "elapsed_seconds": 0.337484297, "prompt_tokens": 29, "cached_tokens": 25, "output_tokens": 3, "first_output_seconds": 0.247440297, "decode_tokens_per_second": 34.326906573602606, "prefill_tokens_per_second": 16.313479828382192 }, { "id": 2, "model": "qwen3.8-27b-ud-q4_k_xl", "state": "done", "started_ms": 1791310703634, "elapsed_seconds": 0.722133893, "prompt_tokens": 29, "cached_tokens": 25, "output_tokens": 3, "first_output_seconds": 0.597317795, "decode_tokens_per_second": 24.646527715020415, "prefill_tokens_per_second": 14.992447554544398 }, { "id": 1, "model": "qwen3.8-27b-ud-q4_k_xl", "state": "done", "started_ms": 1791310675215, "elapsed_seconds": 27.999049124, "prompt_tokens": 29, "cached_tokens": 0, "output_tokens": 3, "first_output_seconds": 0.432180295, "decode_tokens_per_second": 16.981676770764345, "prefill_tokens_per_second": 67.7033844688436 } ], "backends": [ { "backend": "deployment", "instance": "qwen-gpu", "model": "multi-gpu-smoke", "available": true, "counters": {}, "gauges": { "gpu_budget_bytes_estimate": 20110612736.0, "gpu_count": 1.0, "host_budget_bytes_estimate": 24000000000.0, "queue_capacity": 0.0, "slots_available": 1.0 } }, { "backend": "llama.cpp / CUDA", "instance": "llama-55319-http://127.0.0.1:43411", "model": "qwen3.8-27b-ud-q4_k_xl", "available": true, "counters": {}, "gauges": { "active_cached_tokens": 0.0, "active_output_tokens": 0.0, "active_prompt_tokens": 0.0, "context_capacity_tokens": 2048.0, "context_used_tokens": 31.0, "gpu_layers_requested": 999.0, "process_resident_bytes": 1456742400.0, "requests_active": 0.0, "slots_total": 1.0 } } ], "memory": { "observed_at_unix_ms": 1791310703332, "overall_pressure": "hard", "topology": "discrete", "host": { "capacity_bytes": 105469059072, "available_bytes": 99107876864, "managed_bytes": 0, "reserved_bytes": 0, "pressure": "normal" }, "accelerator": { "capacity_bytes": 25769803776, "available_bytes": 3666870272, "managed_bytes": 0, "reserved_bytes": 0, "pressure": "hard" }, "last_action_unix_ms": null, "counters": { "active_reservations": 0, "backend_cleanup_failures": 0, "backend_cleanup_latched": 0, "completed_demotions": 0, "completed_evictions": 0, "failed_releases": 0, "managed_allocations": 0, "orphaned_release_bytes": 0, "pressure_actions_in_flight": 0 } }, "backend_observed_at_ms": 1791310703226 } } ], "diagnostics_after": { "deployments": { "profiles": [ { "available_slots": 1, "plan": { "args": [ "--device", "CUDA0", "--split-mode", "none", "--main-gpu", "0", "--no-warmup" ], "devices": [ { "architecture": "sm_86", "free_bytes": 21701328896, "id": "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9", "name": "NVIDIA GeForce RTX 3090", "numa_node": null, "pci_bus_id": "00000000:03:00.0", "pcie_link": null, "physical_index": 0, "runtime": "cuda", "total_bytes": 25769803776, "vendor": "nvidia", "visible_index": 0 } ], "environment": { "CUDA_DEVICE_ORDER": "PCI_BUS_ID", "CUDA_VISIBLE_DEVICES": "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9", "OMP_NUM_THREADS": "8" }, "executable_sha256": "236d00dc1c71b01bd15a13b229edeac2e4b05a9ac740d33aadc14fe6f71287b6", "fingerprint": "5878014b255b22aa490568d6cf78ad3db30a2c28be70b60414154c5fac06090a", "model_identity": "sha256:1352bb293f26ccf8ee2fd8d226093eff8357e5eac21f078dca9f0e8bc2f6f071", "profile": { "alias": "multi-gpu-smoke", "batch": 256, "build": "ec928150501c", "context": 2048, "cpu_threads": 8, "executable": "/home/philipbodenbach/.local/share/werk1112/backends/llama-cuda/upstream-ec928150501c/build/bin/llama-server", "gpus": [ "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9" ], "host_bytes": 24000000000, "id": "qwen-gpu", "kv_cpu": false, "memory": [ { "compute": 536870912, "experts": 0, "kv": 268435456, "overhead": 268435456, "replicated": 0, "reserve": 536870912, "weights": 18500000000 } ], "model": "qwen3.8-27b-ud-q4_k_xl", "native_args": [ "--no-warmup" ], "offload": { "kind": "none" }, "parallel": 1, "pipeline_parallel": 1, "runtime": "llama_cuda", "split": [], "strategy": "single", "tensor_parallel": 1 }, "reasons": [ "Exclusive GPU ownership; component budgets are declared estimates, not measured usage or hard allocator limits.", "Unknown P2P/NVLink is not treated as a prerequisite; no throughput prediction was made." ], "strategy": "single" }, "queue_capacity": 0, "queue_policy": "reject", "workers": [ { "available": true, "backend": "llama.cpp / CUDA", "counters": {}, "gauges": { "active_cached_tokens": 0.0, "active_output_tokens": 0.0, "active_prompt_tokens": 0.0, "context_capacity_tokens": 2048.0, "context_used_tokens": 31.0, "gpu_layers_requested": 999.0, "process_resident_bytes": 1777864704.0, "requests_active": 0.0, "slots_total": 1.0 }, "instance": "llama-55319-http://127.0.0.1:43411", "model": "qwen3.8-27b-ud-q4_k_xl" } ] } ] }, "inventory": { "devices": [ { "architecture": "sm_86", "free_bytes": 3665821696, "id": "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9", "name": "NVIDIA GeForce RTX 3090", "numa_node": null, "pci_bus_id": "00000000:03:00.0", "pcie_link": null, "physical_index": 0, "runtime": "cuda", "total_bytes": 25769803776, "vendor": "nvidia", "visible_index": 0 } ], "host_available_bytes": 98737479680, "topology": null, "warnings": [] } }, "status": "passed_transport", "correctness": "Outputs retained for reference comparison; transport success alone is not numerical equivalence.", "latency_seconds": { "median": 1.022885390000738, "min": 0.609684594997816, "max": 28.28231289299947 }, "gpu_memory_samples": [ { "time": 1791310671.8969424, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3630 } }, { "time": 1791310672.225934, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3631 } }, { "time": 1791310672.4870093, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3630 } }, { "time": 1791310672.8416502, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3630 } }, { "time": 1791310673.197757, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3630 } }, { "time": 1791310673.5710483, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3627 } }, { "time": 1791310673.8688893, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3627 } }, { "time": 1791310674.1955836, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3627 } }, { "time": 1791310674.4791298, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3611 } }, { "time": 1791310674.8079202, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3611 } }, { "time": 1791310675.1492622, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3611 } }, { "time": 1791310675.443096, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3630 } }, { "time": 1791310675.760475, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3594 } }, { "time": 1791310676.0578268, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3582 } }, { "time": 1791310676.3220139, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3582 } }, { "time": 1791310676.621218, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3582 } }, { "time": 1791310676.922904, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3579 } }, { "time": 1791310677.2035596, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3579 } }, { "time": 1791310677.4833865, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3828 } }, { "time": 1791310677.7914257, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3828 } }, { "time": 1791310678.0940368, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3828 } }, { "time": 1791310678.4024944, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3808 } }, { "time": 1791310678.7108943, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3808 } }, { "time": 1791310679.0195198, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3808 } }, { "time": 1791310679.347412, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3827 } }, { "time": 1791310679.6337242, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3827 } }, { "time": 1791310679.9132907, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3827 } }, { "time": 1791310680.2052531, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3832 } }, { "time": 1791310680.4969625, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3832 } }, { "time": 1791310680.813862, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3832 } }, { "time": 1791310681.1200705, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3832 } }, { "time": 1791310681.423507, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3829 } }, { "time": 1791310681.7087007, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3829 } }, { "time": 1791310682.005145, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3829 } }, { "time": 1791310682.3147168, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3817 } }, { "time": 1791310682.594912, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3808 } }, { "time": 1791310682.912368, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3808 } }, { "time": 1791310683.1892512, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3829 } }, { "time": 1791310683.4984498, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3827 } }, { "time": 1791310683.8298628, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3827 } }, { "time": 1791310684.1298654, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3827 } }, { "time": 1791310684.4265084, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3827 } }, { "time": 1791310684.7078326, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3827 } }, { "time": 1791310685.0110886, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3827 } }, { "time": 1791310685.2994068, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310685.5955067, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3815 } }, { "time": 1791310685.8916354, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3815 } }, { "time": 1791310686.166403, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3815 } }, { "time": 1791310686.4718328, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3798 } }, { "time": 1791310686.7653434, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3798 } }, { "time": 1791310687.0515077, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3798 } }, { "time": 1791310687.3542106, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310687.6472058, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310687.9369843, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310688.2411027, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310688.5463123, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310688.8478608, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310689.1358032, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310689.4517727, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3815 } }, { "time": 1791310689.7327318, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3815 } }, { "time": 1791310690.0316496, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3815 } }, { "time": 1791310690.3278577, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3815 } }, { "time": 1791310690.6243236, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3798 } }, { "time": 1791310690.933005, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3798 } }, { "time": 1791310691.2641292, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3819 } }, { "time": 1791310691.5741293, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310691.8691149, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310692.186784, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310692.487885, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310692.791309, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310693.1209695, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310693.432175, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3806 } }, { "time": 1791310693.7457733, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3806 } }, { "time": 1791310694.0516665, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3806 } }, { "time": 1791310694.3558247, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3798 } }, { "time": 1791310694.6548727, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3765 } }, { "time": 1791310694.9615197, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3765 } }, { "time": 1791310695.2582846, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3819 } }, { "time": 1791310695.5773604, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310695.8786674, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310696.1902933, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310696.4865985, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310696.7907069, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310697.0913463, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 3818 } }, { "time": 1791310697.421734, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 7530 } }, { "time": 1791310697.720855, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 17694 } }, { "time": 1791310698.042727, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 19526 } }, { "time": 1791310698.348709, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 19526 } }, { "time": 1791310698.6549509, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 19485 } }, { "time": 1791310698.9521852, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 19576 } }, { "time": 1791310699.2469497, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 19630 } }, { "time": 1791310699.5440297, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 19629 } }, { "time": 1791310699.8376126, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 19544 } }, { "time": 1791310700.1330543, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 19544 } }, { "time": 1791310700.440118, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 19544 } }, { "time": 1791310700.7313569, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20774 } }, { "time": 1791310701.0290723, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20774 } }, { "time": 1791310701.3358054, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20774 } }, { "time": 1791310701.6319585, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20762 } }, { "time": 1791310701.9587486, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20762 } }, { "time": 1791310702.3046784, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20794 } }, { "time": 1791310702.6156003, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20754 } }, { "time": 1791310702.9082682, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20762 } }, { "time": 1791310703.213853, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20831 } }, { "time": 1791310703.5285246, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20830 } }, { "time": 1791310703.8378172, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20830 } }, { "time": 1791310704.1606662, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20830 } }, { "time": 1791310704.4745414, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20830 } }, { "time": 1791310704.7817285, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20830 } }, { "time": 1791310705.0675876, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20830 } }, { "time": 1791310705.678738, "used_mib": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20822 } } ], "gpu_peak_used_mib_including_other_processes": { "GPU-b41bf955-b7e6-6f64-82b2-6b3cf08450e9": 20831 }, "unavailable_measurements": [ "PCIe transfers", "pinned RAM", "per-worker GPU allocation", "logit/logprob equivalence" ], "installed_llama_revision": "ec928150501c2572fec05cb949061672bb424914" }