hiyouga/LlamaFactory · error · ValueError

ChatGLM model is not supported yet.

Error message

ChatGLM model is not supported yet.

What it means

In the export-time GPTQ branch, after version checks, LlamaFactory explicitly rejects chatglm architectures because optimum/gptqmodel block-pattern handling for ChatGLM is not wired up. The check reads config.model_type == 'chatglm' and raises ValueError.

Source

Thrown at src/llamafactory/model/model_utils/quantization.py:152

            check_version("autoawq", mandatory=True)

        if quant_method == QuantizationMethod.AQLM:
            check_version("aqlm>=1.1.0", mandatory=True)
            quantization_config["bits"] = 2

        quant_bits = quantization_config.get("bits", "?")
        logger.info_rank0(f"Loading {quant_bits}-bit {quant_method.upper()}-quantized model.")

    elif model_args.export_quantization_bit is not None:  # gptqmodel
        if model_args.export_quantization_bit not in [8, 4, 3, 2]:
            raise ValueError("AutoGPTQ only accepts 2/3/4/8-bit quantization.")

        check_version("optimum>=1.24.0", mandatory=True)
        check_version("gptqmodel>=2.0.0", mandatory=True)
        from accelerate.utils import get_max_memory

        if getattr(config, "model_type", None) == "chatglm":
            raise ValueError("ChatGLM model is not supported yet.")

        try:
            from optimum.gptq import utils as gq_utils

            if "language_model.model.layers" not in gq_utils.BLOCK_PATTERNS:
                gq_utils.BLOCK_PATTERNS.insert(0, "language_model.model.layers")
        except ImportError:
            pass

        block_name_to_quantize = None
        if getattr(config, "model_type", None) in ["gemma3", "paligemma"]:
            block_name_to_quantize = "language_model.model.layers"

        init_kwargs["quantization_config"] = GPTQConfig(
            bits=model_args.export_quantization_bit,
            tokenizer=tokenizer,
            dataset=_get_quantization_dataset(tokenizer, model_args),
            block_name_to_quantize=block_name_to_quantize,

View on GitHub (pinned to f28afaf635)

Solutions

1. Export ChatGLM unquantized (drop export_quantization_bit) and quantize with a ChatGLM-aware toolchain, if any supports it.
2. Switch to a supported architecture (llama-family, qwen, etc.) for GPTQ export.
3. Keep ChatGLM in AWQ/other PTQ form from upstream providers rather than quantizing locally via this path.

Example fix

# before (export yaml, chatglm base)
model_name_or_path: THUDM/chatglm3-6b
export_quantization_bit: 4

# after
model_name_or_path: THUDM/chatglm3-6b
# export_quantization_bit removed (export unquantized)

if export_quantization_bit is not None:
    assert getattr(config, "model_type", None) != "chatglm", "GPTQ export does not support chatglm"

def gptq_export_supported(config) -> bool:
    return getattr(config, "model_type", None) != "chatglm"

When it happens

Trigger: Running llamafactory-cli export with export_quantization_bit set on a ChatGLM checkpoint (THUDM/chatglm2-6b, chatglm3-6b, etc.).

Common situations: Attempting to produce GPTQ INT4 exports of ChatGLM2/3 for vLLM-style deployment; reusing a working export config from Llama with a ChatGLM base.

Related errors

• Cannot find satisfying example, considering decrease `export
• AutoGPTQ only accepts 2/3/4/8-bit quantization.
• Invalid API key.
• Invalid length
• Dict is not supported.

AI-assisted analysis of hiyouga/LlamaFactory@f28afaf635 (2026-08-14). Data as JSON: /api/errors/8ecd464af437a727. Report an issue: GitHub.