bert4torch_config / Qwen-14B /bert4torch_config.json
Tongjilibo's picture
为大模型增加generation_config
16f61c2
raw
history blame
No virus
868 Bytes
{
"model": "qwen",
"hidden_act": "silu",
"intermediate_size": 13696,
"initializer_range": 0.02,
"layer_norm_eps": 1e-06,
"hidden_size": 5120,
"num_attention_heads": 40,
"num_hidden_layers": 40,
"rotary_emb_base": 10000,
"rotary_pct": 1.0,
"scale_attn_weights": true,
"seq_length": 2048,
"tie_word_embeddings": false,
"use_cache": true,
"flash_attention": "flash_attn_2",
"vocab_size": 152064,
"use_dynamic_ntk": true,
"use_logn_attn": true,
"segment_vocab_size": 0,
"skip_init": true,
"rope_rank": "updown",
"max_position_embeddings": 8192,
"generation_config": {"tokenizer_config": {"allowed_special": ["<|im_end|>", "<|im_start|>", "<|endoftext|>"],
"skip_special_tokens": true}, "end_id": [151643],
"max_length": 8192}
}