| { | |
| "model": "glm2", | |
| "hidden_act": "swiglu", | |
| "hidden_size": 4096, | |
| "intermediate_size": 13696, | |
| "layer_norm_eps": 1e-05, | |
| "max_sequence_length": 32768, | |
| "num_attention_heads": 32, | |
| "num_hidden_layers": 28, | |
| "vocab_size": 65024, | |
| "segment_vocab_size": 0, | |
| "num_key_value_heads": 2, | |
| "skip_init": true, | |
| "tie_word_embeddings": false, | |
| "eos_token_id": 2, | |
| "pad_token_id": 2, | |
| "rmsnorm": true, | |
| "rope_rank": "adjacent", | |
| "position_encoding_2d": true, | |
| "rope_scaling": {"type": "glm"}, | |
| "torch_dtype": "float16", | |
| "_attn_implementation": "sdpa", | |
| "quantization_bit": 4, | |
| "quantization_method": "cpm_kernels", | |
| "target_modules": ["q", "k", "v", "o", "intermediateDense", "outputDense"], | |
| "generation_config": {"max_length": 32768} | |
| } |