bert4torch_config / Llama-2-13b-hf /bert4torch_config.json
Tongjilibo's picture
为大模型增加generation_config
16f61c2
raw history blame
No virus
552 Bytes
{
"model": "llama",
"bos_token_id": 1,
"eos_token_id": 2,
"pad_token_id": 0,
"hidden_size": 5120,
"intermediate_size": 13824,
"num_attention_heads": 40,
"num_hidden_layers": 40,
"initializer_range": 0.02,
"hidden_act": "silu",
"vocab_size": 32000,
"segment_vocab_size": 0,
"skip_init": true,
"layer_norm_eps": 1e-5,
"rope_rank": "updown",
"tie_word_embeddings": false,
"torch_dtype": "float16",
"generation_config": {"tokenizer_config": {"skip_special_tokens": true, "add_special_tokens": false},"end_id": 2}
}