[quantization] Add per-token NVFP4 quantization MoE support for ReLU2 (#56740)

Signed-off-by: Xuanteng Huang <xuantengh@nvidia.com>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>
This commit is contained in:
Xuanteng Huang
2026-10-07 22:01:09 +00:00
committed by GitHub
co-authored by Claude Fable 5.1 OpenAI Codex
parent b803edbc9f
commit f38c6778ce
18 changed files with 450 additions and 61 deletions
+5
View File
@@ -110,6 +110,11 @@ class DeepseekV4FP8Config(Fp8Config):
kv_cache_quant_algo=None,
exclude_modules=[],
group_size=16,
quantization_args=(
self.online_quantization_config.args
if self.online_quantization_config is not None
else None
),
)
return self._nvfp4_config
+5
View File
@@ -114,6 +114,11 @@ class DeepseekV4FP8Config(Fp8Config):
kv_cache_quant_algo=None,
exclude_modules=[],
group_size=16,
quantization_args=(
self.online_quantization_config.args
if self.online_quantization_config is not None
else None
),
)
return self._nvfp4_config