Skip to content

[CI] remove lora_bias_stacked as it is deprecated in vllm #835

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

Open
wants to merge 3 commits into
base: main
Choose a base branch
from
Open

[CI] remove lora_bias_stacked as it is deprecated in vllm #835

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
8497100
[CI] remove lora_bias_stacked as i has been deprecated in vllm
Oct 11, 2025
aa23fbb
fix presubmit error
Oct 11, 2025
fe83ece
update interface of original_set_lora based on vllm side change
Oct 12, 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
7 changes: 0 additions & 7 deletions tpu_inference/layers/vllm/sharding.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -102,22 +102,15 @@ def _shard_base_linear_lora(layer: BaseLinearLayerWithLoRA,
# NOTE: lora_a_stacked[i] has shape [max_loras, 1, num_out, num_in]
sharded_lora_a_tpu = torch.nn.ParameterList()
sharded_lora_b_tpu = torch.nn.ParameterList()
sharded_lora_bias_tpu = torch.nn.ParameterList()

for i in range(layer.n_slices):
sharded_lora_a_tpu.append(
_shard_tensor_to_tpu_replicated(layer.lora_a_stacked[i], mesh))
sharded_lora_b_tpu.append(
_shard_tensor_to_tpu_replicated(layer.lora_b_stacked[i], mesh))
if layer.lora_bias_stacked is not None:
sharded_lora_bias_tpu.append(
_shard_tensor_to_tpu_replicated(layer.lora_bias_stacked[i],
mesh))

layer.lora_a_stacked = sharded_lora_a_tpu
layer.lora_b_stacked = sharded_lora_b_tpu
if layer.lora_bias_stacked is not None:
layer.lora_bias_stacked = sharded_lora_bias_tpu


# TODO: Add custom sharding logic for following lora layers
Expand Down
4 changes: 1 addition & 3 deletions tpu_inference/models/vllm/vllm_model_wrapper.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -255,11 +255,9 @@ def _tpu_set_lora(
lora_a: torch.Tensor,
lora_b: torch.Tensor,
embeddings_tensor: Optional[torch.Tensor],
bias: Optional[torch.Tensor] = None,
):
with torchax.default_env():
self._original_set_lora(index, lora_a, lora_b, embeddings_tensor,
bias)
self._original_set_lora(index, lora_a, lora_b, embeddings_tensor)

def _tpu_reset_lora(self, index: int):
with torchax.default_env():
Expand Down
Loading