Skip to content

[Model] Qwen3 Dense FP8 Compat Fixes #17318

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 2 commits into from
Apr 28, 2025
Merged

[Model] Qwen3 Dense FP8 Compat Fixes #17318

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a647d99
[Model] Qwen3 Dense FP8 Compat Fixes
simon-mo Apr 28, 2025
562f460
lint
simon-mo Apr 28, 2025
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
9 changes: 9 additions & 0 deletions vllm/model_executor/layers/linear.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -929,6 +929,15 @@ def weight_loader_v2(self,
shard_offset = self._get_shard_offset_mapping(loaded_shard_id)
shard_size = self._get_shard_size_mapping(loaded_shard_id)

# Note(simon): This is needed for Qwen3's fp8 quantization.
if isinstance(param, BlockQuantScaleParameter):
assert self.quant_method is not None
assert hasattr(self.quant_method, "quant_config")
weight_block_size = self.quant_method.quant_config.weight_block_size
block_n, _ = weight_block_size[0], weight_block_size[1]
shard_offset = (shard_offset + block_n - 1) // block_n
shard_size = (shard_size + block_n - 1) // block_n

param.load_qkv_weight(loaded_weight=loaded_weight,
num_heads=self.num_kv_head_replicas,
shard_id=loaded_shard_id,
Expand Down