[Core] Support fp32 lm_head for generation models via head_dtype (RFC #48305 §3.6) (#48390)

Signed-off-by: Karthik Kothuri <karthikkothuri2009@gmail.com>
Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: wang.yuqi <yuqi.wang@daocloud.io>
This commit is contained in:
Karthik Kothuri
2026-07-13 16:43:34 +08:00
committed by GitHub
co-authored by Claude wang.yuqi
parent 56a357ed33
commit 107a03ba63
6 changed files with 249 additions and 15 deletions
@@ -126,3 +126,4 @@ def wikitext_ppl_test(
# We are not concerned that the vllm PPL is less than Transformers,
# so we only perform one-sided testing.
assert differ < atol
return vllm_ppl