config.json

{
  "BATCH_ADAPTIVE_INIT_POS": false,
  "BiPE_seps": [
    15,
    13,
    32,
    2,
    28,
    27,
    209,
    186,
    187
  ],
  "EXCLUDE_DIAGONAL": true,
  "PADDING_ID": 0,
  "PRINT_KV_RATIO": true,
  "SA_Denominator_Bias": 1e-10,
  "SA_Numerator_Bias": 0.0,
  "USE_BiPE": true,
  "USE_GENERATE_LOCAL_WIN_SIZES_wrt_LAYERS": false,
  "USE_ORIGINAL_FULL_ATTEN": false,
  "USE_PREFILL_LOCAL_WIN_SIZES_wrt_LAYERS": false,
  "USE_SA_SOFTMAX": false,
  "USE_SA_SOFTMAX_NO_DENO": false,
  "architectures": [
    "SepLLMGPTNeoXForCausalLM"
  ],
  "attention_bias": true,
  "attention_dropout": 0.0,
  "bos_token_id": 0,
  "classifier_dropout": 0.1,
  "eos_token_id": 2,
  "generate_k": 0,
  "generate_local_window_size": 64,
  "generate_win_loc_size_list": [
    2048,
    256,
    256,
    256,
    256,
    2048,
    256,
    256,
    256,
    256,
    256,
    2048
  ],
  "hidden_act": "gelu",
  "hidden_dropout": 0.0,
  "hidden_size": 768,
  "init_tok_max_idx": 2,
  "initializer_range": 0.02,
  "intermediate_size": 3072,
  "layer_norm_eps": 1e-05,
  "max_position_embeddings": 2048,
  "model_parallel_size": 1,
  "model_type": "sepllm_gpt_neox",
  "mup_rp_embedding_mult": 1.0,
  "num_attention_heads": 12,
  "num_hidden_layers": 12,
  "num_layers": 12,
  "pos_emb": "rotary",
  "prefill_k": 0,
  "prefill_loc_win_size_list": [
    2048,
    256,
    256,
    256,
    256,
    2048,
    256,
    256,
    256,
    256,
    256,
    2048
  ],
  "prefill_local_window_size": 64,
  "print_ratio_intervals": 100,
  "rope_scaling": null,
  "rotary_emb_base": 10000.0,
  "rotary_pct": 0.25,
  "separator_token_ids": [
    15,
    13,
    32,
    2,
    28,
    27,
    209,
    186,
    187
  ],
  "streamingLLM": false,
  "tie_word_embeddings": false,
  "torch_dtype": "float16",
  "transformers_version": "4.38.0",
  "use_cache": true,
  "use_parallel_residual": true,
  "vocab_size": 50304
}