{
  "model": "Qwen/Qwen3.5-2B",
  "output_repo": "saicr/Full-Model-Background-Processing",
  "total_sequences": 1000000,
  "max_input_tokens": 128,
  "min_input_tokens_target": 48,
  "max_new_tokens": 16,
  "batch_size": 16,
  "shard_sequences": 128,
  "datasets": [
    {
      "short_name": "ultrachat",
      "repo": "HuggingFaceH4/ultrachat_200k",
      "config": null,
      "split": "train_sft",
      "field_mode": "prompt"
    },
    {
      "short_name": "fineweb_edu",
      "repo": "HuggingFaceFW/fineweb-edu",
      "config": "sample-10BT",
      "split": "train",
      "field_mode": "text"
    },
    {
      "short_name": "openwebmath",
      "repo": "open-web-math/open-web-math",
      "config": null,
      "split": "train",
      "field_mode": "text"
    },
    {
      "short_name": "cosmopedia_v2",
      "repo": "HuggingFaceTB/smollm-corpus",
      "config": "cosmopedia-v2",
      "split": "train",
      "field_mode": "cosmopedia"
    }
  ],
  "formats": [
    "raw",
    "qa_short",
    "user_assistant",
    "hash_sections",
    "xml_document",
    "key_point",
    "continue",
    "next_sentence",
    "context_answer",
    "brackets",
    "reader_reply",
    "plain_instruction",
    "title",
    "summary",
    "minimal",
    "note",
    "completion",
    "quoted",
    "dash",
    "task_block"
  ],
  "sampling": {
    "temperature": 0.85,
    "top_k": 50,
    "top_p": 0.95
  },
  "capture": {
    "meaning": "Transformers hidden_states tuple: embedding/residual stream plus all model layers",
    "hidden_size": 2048,
    "num_hidden_layers": 24,
    "num_hidden_state_tensors": 25,
    "prefill": true,
    "generated_tokens": true
  },
  "quantization": {
    "dtype": "int8",
    "scheme": "symmetric per-token per-hidden-state",
    "scale_dtype": "float16",
    "formula": "float ~= int8 * scale"
  },
  "tokenizer_class": "Qwen2Tokenizer",
  "created_unix": 1792240935.4077983
}