{
  "base_model": "Qwen/Qwen3.6-27B",
  "data_dir": "data/processed/level4_annotated_translation_2026-05-06",
  "seq_len": 2048,
  "batch_size": 2,
  "grad_accum": 8,
  "effective_batch_sequences": 16,
  "cuda_available": true,
  "gpu_name": "NVIDIA H200 NVL",
  "gpu_memory_gib": 139.81,
  "libraries": {
    "torch": "2.11.0+cu128",
    "transformers": "5.7.0",
    "peft": "0.19.1",
    "datasets": "4.8.5",
    "accelerate": "1.13.0",
    "triton": "3.6.0",
    "einops": "0.8.2",
    "fla": "0.5.0",
    "tilelang": "0.1.9",
    "causal_conv1d": null,
    "flash_attn": null
  },
  "cuda_home": "/home/jaysuh2/joseon-to-day/.cache/cuda-12.8",
  "tilelang_target": "target.Target(kind=target.TargetKind(name=\"cuda\", default_device_type=2, default_keys=(\"cuda\", \"gpu\")), tag=\"\", keys=(\"cuda\", \"gpu\"), attrs={\"thread_warp_size\": 32, \"max_num_threads\": 1024, \"arch\": \"sm_90a\"}, features={}, host=None)",
  "tilelang_target_error": null,
  "missing_data": [],
  "notes": [
    "causal-conv1d is missing; Transformers will warn that the full Qwen3.6 fast path is unavailable.",
    "flash-attn is missing; full-attention layers will use SDPA instead of FlashAttention2."
  ]
}