[model] support MiMo-V2-Flash model (#9637)

2026-01-31 06:42:05 +00:00 · 2025-12-21 14:38:18 +08:00
parent 0894b4f37e
commit 4923f52a28
5 changed files with 39 additions and 1 deletions
--- a/src/llamafactory/model/model_utils/quantization.py
+++ b/src/llamafactory/model/model_utils/quantization.py
@@ -113,6 +113,7 @@ def configure_quantization(
        if quant_method == QuantizationMethod.FP8:
            quant_config = FineGrainedFP8Config(dequantize=True)
            init_kwargs["quantization_config"] = quant_config
+            init_kwargs["ignore_mismatched_sizes"] = True

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