== residency proof (card t_d199e09c) ==
date: 2026-09-22T16:30:22Z
model: /work/t_d199e09c/models/MiMo-V2.6-Distill-Qwen-9B-Q4_K_M.gguf
size: 5841049120 bytes
sha256: 4bca6f18c73f72270c7a20c2ea2bea581de8246e318714277120369d34048c81

-- nvidia-smi before the load (desktop/voice-app floor already counted) --
name, memory.total [MiB], memory.used [MiB], memory.free [MiB]
NVIDIA GeForce RTX 3060 Ti, 8192 MiB, 1038 MiB, 6854 MiB

-- fit plan for the placement this load will use --
cmd: uv run typed-gguf fit /work/t_d199e09c/models/MiMo-V2.6-Distill-Qwen-9B-Q4_K_M.gguf --print --json --no-cache --n-ctx 4096 --fit-ctx 4096 --fit-target 512
{
 "n_gpu_layers": 32,
 "n_ctx": 4096,
 "kv_type": "q4_0",
 "n_seq_max": 8,
 "est_weights_bytes": 5829033984,
 "est_kv_bytes": 150994944,
 "est_total_bytes": 6534725632,
 "budget_bytes": 6650068992,
 "source": "llama-fit-params",
 "warnings": [
  "W_KV_TYPE_DOWNGRADE"
 ],
 "notes": [
  "memory table from b11026-linux-x64-vulkan/llama-fit-params (model 5559 MiB, context 177 MiB, compute 529 MiB)"
 ]
}

-- one real request through the driver's exact invocation --
cmd: uv run typed-gguf run --questions /work/t_d199e09c/mimo/questions.json --state @/work/t_d199e09c/mimo/items/t07.txt --model /work/t_d199e09c/models/MiMo-V2.6-Distill-Qwen-9B-Q4_K_M.gguf --threads 4 --fit-target 512 --out /work/t_d199e09c/mimo/proof/t07.json --keep-alive 1m
exit=0 wall_ms=7884

-- nvidia-smi while the model is resident --
memory.total [MiB], memory.used [MiB], memory.free [MiB]
8192 MiB, 5907 MiB, 1985 MiB

-- keep host's own placement block (from keep status --json) --
{
 "model": "MiMo-V2.6-Distill-Qwen-9B-Q4_K_M",
 "model_path": "/work/t_d199e09c/models/MiMo-V2.6-Distill-Qwen-9B-Q4_K_M.gguf",
 "state": "running",
 "pid": 69876,
 "requests": 1,
 "model_load_ms": 1449.3314270002884,
 "placement": {
  "note": "fit plan: 32 layer(s) offloaded, kv_type=q4_0",
  "n_gpu_layers": 32,
  "kv_type": "q4_0",
  "degraded": false,
  "attempts": [],
  "warnings": [],
  "cpu_only": false
 },
 "devices": {
  "devices": [
   "CPU",
   "CPU_Mapped",
   "Vulkan0",
   "Vulkan_Host"
  ],
  "device_buffers": {
   "Vulkan0": 1,
   "Vulkan_Host": 1
  },
  "effective_backend": "vulkan"
 },
 "key": {
  "model_path": "/work/t_d199e09c/models/MiMo-V2.6-Distill-Qwen-9B-Q4_K_M.gguf",
  "model_sha": "stat:5841049120:1790094309194734270",
  "backend": "auto",
  "n_ctx": null,
  "n_seq_max": null,
  "kv_type": "auto",
  "threads": 4,
  "fit": true,
  "fit_target_mb": 512,
  "fit_ctx": null,
  "fit_cache": true
 },
 "loaded_s": 1.185
}

-- the response's own engine block (from the payload it wrote) --
{
 "engine.runtime": "llama.cpp b11026",
 "engine.backend": "vulkan",
 "engine.effective_backend": "vulkan",
 "engine.devices": [
  "CPU",
  "CPU_Mapped",
  "Vulkan0",
  "Vulkan_Host"
 ],
 "engine.readout": "sequence",
 "engine.cue": "json_instructed",
 "engine.chat_format": {
  "kind": "role_split",
  "question_turn": "user",
  "contract": "question",
  "prefix_chars": 438,
  "dropped": ""
 },
 "engine.template": {
  "kind": "builtin",
  "renderer": "builtin",
  "source": "llama_chat_apply_template",
  "family": "qwen35",
  "thinking": "suppressed",
  "warnings": [
   "W_TEMPLATE_FALLBACK"
  ],
  "notes": [
   "the internal renderer rejected this template; the runtime's built-in family table rendered it"
  ],
  "explicit": false
 },
 "engine.kv_unified": true,
 "engine.n_ctx": 256,
 "engine.n_seq_max": 8,
 "engine.kv_type": "q4_0",
 "engine.n_gpu_layers": 32,
 "engine.placement": {
  "note": "fit plan: 32 layer(s) offloaded, kv_type=q4_0",
  "n_gpu_layers": 32,
  "kv_type": "q4_0",
  "degraded": false,
  "attempts": [],
  "warnings": [],
  "cpu_only": false
 },
 "engine.fit": {
  "n_gpu_layers": 32,
  "n_ctx": 4096,
  "kv_type": "q4_0",
  "n_seq_max": 8,
  "est_weights_bytes": 5829033984,
  "est_kv_bytes": 150994944,
  "est_total_bytes": 6534725632,
  "backend": "cpu",
  "source": "llama-fit-params",
  "warnings": [
   "W_KV_TYPE_DOWNGRADE"
  ],
  "notes": [
   "memory table from b11026-linux-x64-vulkan/llama-fit-params (model 5559 MiB, context 177 MiB, compute 529 MiB)"
  ],
  "arch": "qwen35",
  "model_sha256": "4bca6f18c73f72270c7a20c2ea2bea581de8246e318714277120369d34048c81",
  "host_fingerprint": "a8d49ce94b68e6f8",
  "budget_bytes": 6647971840,
  "created_at": "2026-09-22T16:26:14Z",
  "schema": "typed_gguf.fit/v1"
 },
 "engine.keep": {
  "served_by": "host",
  "keep_alive_s": 60.0,
  "state": "running",
  "pid": 69876,
  "key": {
   "model_path": "/work/t_d199e09c/models/MiMo-V2.6-Distill-Qwen-9B-Q4_K_M.gguf",
   "model_sha": "stat:5841049120:1790094309194734270",
   "backend": "auto",
   "n_ctx": null,
   "n_seq_max": null,
   "kv_type": "auto",
   "threads": 4,
   "fit": true,
   "fit_target_mb": 512,
   "fit_ctx": null,
   "fit_cache": true
  },
  "key_digest": "f6df7f6d93d238eb",
  "model": "MiMo-V2.6-Distill-Qwen-9B-Q4_K_M",
  "model_path": "/work/t_d199e09c/models/MiMo-V2.6-Distill-Qwen-9B-Q4_K_M.gguf",
  "requests": 1,
  "uptime_s": 7.631,
  "idle_left_s": 59.003,
  "loaded_at": 1790094630.9012294,
  "placement": {
   "note": "fit plan: 32 layer(s) offloaded, kv_type=q4_0",
   "n_gpu_layers": 32,
   "kv_type": "q4_0",
   "degraded": false,
   "attempts": [],
   "warnings": [],
   "cpu_only": false
  },
  "devices": {
   "devices": [
    "CPU",
    "CPU_Mapped",
    "Vulkan0",
    "Vulkan_Host"
   ],
   "device_buffers": {
    "Vulkan0": 1,
    "Vulkan_Host": 1
   },
   "effective_backend": "vulkan"
  },
  "model_load_ms": 1449.3314270002884,
  "socket": "/work/t_d199e09c/mimo/home/keep/f6df7f6d93d238eb.sock",
  "spec": "/work/t_d199e09c/mimo/home/keep/f6df7f6d93d238eb.spec.json",
  "log": "/work/t_d199e09c/mimo/home/keep/f6df7f6d93d238eb.log",
  "version": "",
  "fallback": null
 },
 "engine.device_buffers": {
  "Vulkan0": 1,
  "Vulkan_Host": 1
 }
}

-- keep host stopped --
{
  "stopped": true,
  "pid": 69876,
  "reason": "stopped (SIGTERM)",
  "cleaned": false
}
memory.total [MiB], memory.used [MiB], memory.free [MiB]
8192 MiB, 1033 MiB, 6858 MiB
