Skip to content

🐛 Fix fp8 model name check with quantization check #535

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Oct 17, 2025
Merged

🐛 Fix fp8 model name check with quantization check #535

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion vllm_spyre/v1/worker/spyre_worker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -444,7 +444,8 @@ def _warmup_spyre_dynamic_size(self, special_token_ids):
0, len(valid_token_ids_tensor), (3, prompt_len))]

# TODO: we need 2 requests for warmup on FP8+CB
is_fp8_plus_cb = 'FP8' in self.model_config.model and \
# Check if model is quantized
Copy link
Collaborator

@ckadner ckadner Oct 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What about other quantizations like 4 bit or 8 bit int?

There is some more code from @joerunde and some I recently wrote to do this kind of check, which could be adapted and used here:

Copy link
Collaborator

@joerunde joerunde Oct 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We'll have to refactor all these checks once we support more quantization methods, today we only support fp8. I wouldn't mind a refactor to at least pull all of these fp8 checks into one helper instance method in the model class, but we don't need to block this fix on it

I do think this is a separate problem than figuring out which model we're serving though, because any fp8 model has to be handled separately here, not just granite specifically.

Copy link
Collaborator

@ckadner ckadner Oct 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Created a follow-up issue: #537

is_fp8_plus_cb = self.model_config.quantization is not None and \
envs_spyre.VLLM_SPYRE_USE_CB
req_count = 3 if is_fp8_plus_cb else 2
requests = [
Expand Down