[Quantization] Support compressed-tensors MXFP4 MoE with FP8 activations via Mxfp4MoEMethod (#59815)

Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Kevin H. Luu <khluu000@gmail.com>
This commit is contained in:
Misha Goin
2026-10-14 13:33:35 -07:00
committed by GitHub
co-authored by Claude Kevin H. Luu
parent fe98d90a9d
commit 984c242ede
12 changed files with 311 additions and 10 deletions
+1
View File
@@ -752,6 +752,7 @@ class DeepseekV4MoE(nn.Module):
renormalize=config.norm_topk_prob,
quant_config=quant_config,
prefix=f"{prefix}.experts",
ckpt_names=("w1", "w2", "w3"),
scoring_func=self.scoring_func,
routed_scaling_factor=self.routed_scaling_factor,
e_score_correction_bias=self.gate.e_score_correction_bias,