Standardise weight tying on ParallelLMHead.tie_weights (#52147)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Harry Mellor
2026-08-13 15:10:09 +00:00
committed by GitHub
co-authored by Claude Opus 5
parent 7553aac77b
commit 80d6d557f3
56 changed files with 191 additions and 214 deletions
+1 -1
View File
@@ -305,7 +305,7 @@ class OrionForCausalLM(nn.Module, SupportsPP):
prefix=maybe_prefix(prefix, "lm_head"),
)
if self.config.tie_word_embeddings:
self.lm_head.weight = self.model.embed_tokens.weight
self.lm_head = self.lm_head.tie_weights(self.model.embed_tokens)
self.logits_processor = LogitsProcessor(config.vocab_size)
self.make_empty_intermediate_tensors = (
self.model.make_empty_intermediate_tensors