config.json

{
  "_name_or_path": "adowu/astral-128k",
  "architectures": [
    "MistralForCausalLM"
  ],
  "attention_dropout": 0.1,
  "bos_token_id": 1,
  "eos_token_id": 2,
  "hidden_act": "silu",
  "hidden_size": 4096,  // Adjusted to match the checkpoint's expected size
  "initializer_range": 0.02,
  "intermediate_size": 16384,  // Adjusted, assuming the checkpoint expects larger intermediate sizes
  "max_position_embeddings": 128000,  // Kept as is, assuming no mismatch here
  "model_type": "mistral",
  "num_attention_heads": 16,  // Adjusted based on typical configurations for this hidden_size; may need to match checkpoint
  "num_hidden_layers": 16,  // Assuming no mismatch here, kept as is
  "num_key_value_heads": 8,  // This typically doesn't need to change, but ensure it matches the checkpoint if specified
  "pad_token_id": 2,
  "rms_norm_eps": 1e-05,
  "rope_theta": 10000.0,
  "sliding_window": 4096,  // Kept as is, assuming no mismatch here
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",  // Suitable for models running on TPUs or requiring reduced precision for speed
  "transformers_version": "4.39.3",
  "use_cache": true,
  "vocab_size": 32002  // Assuming no mismatch here, kept as is
}