Add LM head quantization support for ModelOpt (#42124)

Signed-off-by: weimingc <17592131+meenchen@users.noreply.github.com>
This commit is contained in:
Wei-Ming Chen
2026-05-26 09:21:05 -07:00
committed by GitHub
parent c8414a8271
commit 6f5b533241
8 changed files with 220 additions and 5 deletions
+1
View File
@@ -875,6 +875,7 @@ class NemotronHForCausalLM(
self.lm_head = ParallelLMHead(
config.vocab_size,
config.hidden_size,
quant_config=self.quant_config,
prefix=maybe_prefix(prefix, "lm_head"),
)
+1
View File
@@ -477,6 +477,7 @@ class Qwen3_5ForCausalLMBase(
self.lm_head = ParallelLMHead(
config.vocab_size,
config.hidden_size,
quant_config=self.quant_config,
prefix=maybe_prefix(prefix, "lm_head"),
)
else:
@@ -381,6 +381,7 @@ class Qwen3_5MTP(nn.Module, SupportsMultiModal):
self.lm_head = ParallelLMHead(
config.vocab_size,
config.hidden_size,
quant_config=self.quant_config,
prefix=maybe_prefix(prefix, "lm_head"),
)
else: