bert4torch_config / Qwen-1_8B /bert4torch_config.json
Tongjilibo's picture
为大模型增加generation_config
16f61c2
{
"model": "qwen",
"hidden_act": "silu",
"bias_dropout_fusion": true,
"bos_token_id": 151643,
"embd_pdrop": 0.1,
"eos_token_id": 151643,
"intermediate_size": 5504,
"initializer_range": 0.02,
"kv_channels": 128,
"layer_norm_eps": 1e-06,
"hidden_size": 2048,
"num_attention_heads": 16,
"num_hidden_layers": 24,
"n_positions": 6144,
"resid_pdrop": 0.1,
"rotary_emb_base": 10000,
"rotary_pct": 1.0,
"scale_attn_weights": true,
"seq_length": 8192,
"tie_word_embeddings": false,
"use_cache": true,
"use_flash_attn": true,
"vocab_size": 151936,
"use_dynamic_ntk": true,
"use_logn_attn": true,
"segment_vocab_size": 0,
"skip_init": true,
"rope_rank": "updown",
"max_position": 8192,
"generation_config": {"tokenizer_config": {"allowed_special": ["<|im_end|>", "<|im_start|>", "<|endoftext|>"],
"skip_special_tokens": true}, "end_id": [151643],
"max_length": 8192}
}