config.json

{
    "_name_or_path": "proteinglm-3b-mlm",
    "add_bias_linear": true,
    "add_qkv_bias": true,
    "apply_query_key_layer_scaling": true,
    "apply_residual_connection_post_layernorm": true,
    "architectures": [
      "ProteinGLMModel"
    ],
    "attention_dropout": 0.0,
    "attention_softmax_in_fp32": true,
    "auto_map": {
      "AutoConfig": "configuration_proteinglm.ProteinGLMConfig",
      "AutoModel": "modeling_proteinglm.ProteinGLMForMaskedLM",
      "AutoModelForCausalLM": "modeling_proteinglm.ProteinGLMForCasualLM",
      "AutoModelForMaskedLM": "modeling_proteinglm.ProteinGLMForMaskedLM",
      "AutoModelForSequenceClassification": "modeling_proteinglm.ProteinGLMForSequenceClassification",
      "AutoModelForTokenClassification": "modeling_proteinglm.ProteinGLMForTokenClassification"
    },
    "bias_dropout_fusion": true,
    "deepnorm": true,
    "experts_per_token": 0,
    "ffn_hidden_size": 6832,
    "fp32_residual_connection": false,
    "glu_activation": "geglu",
    "initializer_range": 0.02, 
    "head_num": 1,
    "hidden_dropout": 0.0,
    "hidden_size": 2560,
    "is_causal": false,
    "use_cache": true,
    "kv_channels": 64,
    "layernorm_epsilon": 1e-05,
    "model_type": "ProteinGLM",
    "moe": false,
    "multi_query_attention": false,
    "multi_query_group_num": 1,
    "num_attention_heads": 40,
    "num_experts": 0,
    "num_layers": 36,
    "padded_vocab_size": 128,
    "post_layer_norm": true,
    "quantization_bit": 0,
    "rmsnorm": false,
    "rotary_embedding_2d": false,
    "seq_length": 2048,
    "torch_dtype": "float32",
    "transformers_version": "4.41.2",
    "untie_head": false,
    "use_pytorch_sdpa": true,
    "vocab_size": 128
  }