config.json

{
    "adapter_type": "mlp",
    "architectures": [
        "GlmasrModel"
    ],
    "attn_implementation": "flash_attention_2",
    "auto_map": {
        "AutoConfig": "configuration_glmasr.GlmasrConfig",
        "AutoModelForCausalLM": "modeling_glmasr.GlmasrModel"
    },
    "max_length": 65536,
    "max_whisper_length": 1500,
    "merge_factor": 4,
    "mlp_adapter_act": "gelu",
    "model_type": "glmasr",
    "quantization": {
        "group_size": 64,
        "bits": 4,
        "mode": "affine"
    },
    "quantization_config": {
        "group_size": 64,
        "bits": 4,
        "mode": "affine"
    },
    "torch_dtype": "bfloat16",
    "transformers_version": "4.51.3",
    "use_rope": true
}