Add support for tie_word_embeddings when loading weights + support for SmolLM #1508
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Motivation
When trying to add support for SmolLM, I noticed that its config dict has
"tie_word_embeddings": true
which isn't supported by SGLang, and as a result,lm_head.weight
is not loaded. This problem likely exists for some other models too (not just SmolLM).Modifications
Within
LlamaForCausalLM
, at the end of the weight loading stage, I added the following code:This can probably be copy-pasted to other LM classes other than
LlamaForCausalLM
, but I'm less familiar with those, and will defer to maintainers to decide whether to do the copying.The effect can be tested by running
python3 -m sglang.bench_latency --correct --model HuggingFaceTB/SmolLM-135M-Instruct
. This would output gibberish before the change, but output coherent answers after it.Checklist