config.json

{
  "architectures": [
    "Glm4ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "eos_token_id": [
    151329,
    151336,
    151338
  ],
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 6144,
  "initializer_range": 0.02,
  "intermediate_size": 23040,
  "max_position_embeddings": 32768,
  "model_type": "glm4",
  "num_attention_heads": 48,
  "num_hidden_layers": 61,
  "num_key_value_heads": 2,
  "pad_token_id": 151329,
  "partial_rotary_factor": 0.5,
  "quantization_config": {
    "act_bits": 16,
    "act_data_type": "int",
    "act_dynamic": true,
    "act_group_size": 128,
    "act_sym": true,
    "amp": true,
    "autoround_version": "0.5.1",
    "batch_size": 8,
    "bits": 4,
    "damp_percent": 0.01,
    "data_type": "int",
    "desc_act": false,
    "enable_minmax_tuning": true,
    "enable_norm_bias_tuning": false,
    "enable_quanted_input": true,
    "gradient_accumulate_steps": 1,
    "group_size": 128,
    "iters": 256,
    "low_gpu_mem_usage": true,
    "lr": 0.00390625,
    "minmax_lr": 0.00390625,
    "nsamples": 256,
    "quant_method": "gptq",
    "scale_dtype": "torch.float16",
    "seqlen": 4096,
    "super_bits": null,
    "super_group_size": null,
    "sym": true,
    "to_quant_block_names": null,
    "true_sequential": false
  },
  "rms_norm_eps": 1e-05,
  "rope_theta": 10000.0,
  "tie_word_embeddings": false,
  "torch_dtype": "float16",
  "transformers_version": "4.51.3",
  "use_cache": true,
  "vocab_size": 151552
}