tllm_checkpoint_float16/decoder/config.json

{
    "architecture": "DecoderModel",
    "dtype": "float16",
    "logits_dtype": "float16",
    "num_hidden_layers": 16,
    "num_attention_heads": 20,
    "hidden_size": 1280,
    "norm_epsilon": 1e-05,
    "vocab_size": 7832,
    "hidden_act": "gelu",
    "use_parallel_embedding": false,
    "embedding_sharding_dim": 0,
    "max_position_embeddings": 5000,
    "use_prompt_tuning": false,
    "head_size": 64,
    "has_position_embedding": true,
    "layernorm_type": 0,
    "has_attention_qkvo_bias": true,
    "has_mlp_bias": true,
    "has_model_final_layernorm": true,
    "has_embedding_layernorm": false,
    "has_embedding_scale": true,
    "ffn_hidden_size": 5120,
    "q_scaling": 1.0,
    "layernorm_position": 0,
    "relative_attention": false,
    "max_distance": 0,
    "num_buckets": 0,
    "model_type": "whisper",
    "rescale_before_lm_head": false,
    "encoder_hidden_size": 1280,
    "encoder_num_heads": 20,
    "encoder_head_size": null,
    "skip_cross_kv": false,
    "quantization": {
        "quant_algo": null
    }
}