bapatra's picture
Add attention_bias to make TGI work (#3)
1adb635 verified
raw
history blame
1.47 kB
{
"_name_or_path": "Phi-3-small-8k-instruct",
"architectures": [
"Phi3SmallForCausalLM"
],
"attention_dropout_prob": 0.0,
"auto_map": {
"AutoConfig": "configuration_phi3_small.Phi3SmallConfig",
"AutoModelForCausalLM": "modeling_phi3_small.Phi3SmallForCausalLM",
"AutoTokenizer": "tokenization_phi3_small.Phi3SmallTokenizer"
},
"blocksparse_block_size": 64,
"blocksparse_homo_head_pattern": false,
"blocksparse_num_local_blocks": 16,
"blocksparse_triton_kernel_block_size": 64,
"blocksparse_vert_stride": 8,
"bos_token_id": 100257,
"dense_attention_every_n_layers": 2,
"embedding_dropout_prob": 0.1,
"eos_token_id": 100257,
"ff_dim_multiplier": null,
"ff_intermediate_size": 14336,
"ffn_dropout_prob": 0.1,
"gegelu_limit": 20.0,
"gegelu_pad_to_256": true,
"hidden_act": "gegelu",
"hidden_size": 4096,
"initializer_range": 0.02,
"layer_norm_epsilon": 1e-05,
"max_position_embeddings": 8192,
"model_type": "phi3small",
"mup_attn_multiplier": 1.0,
"mup_embedding_multiplier": 10.0,
"mup_use_scaling": true,
"mup_width_multiplier": 8.0,
"num_attention_heads": 32,
"num_hidden_layers": 32,
"num_key_value_heads": 8,
"pad_sequence_to_multiple_of_64": true,
"reorder_and_upcast_attn": false,
"rope_embedding_base": 1000000,
"rope_position_scale": 1.0,
"torch_dtype": "bfloat16",
"transformers_version": "4.38.1",
"use_cache": true,
"attention_bias": false,
"vocab_size": 100352
}