Skip to content

Commit

Permalink
Update requirements-tpu.txt to support python 3.9 and 3.11 (#11695)
Browse files Browse the repository at this point in the history
Signed-off-by: mgoin <[email protected]>
  • Loading branch information
mgoin authored Jan 3, 2025
1 parent a655eb3 commit bf0d97d
Show file tree
Hide file tree
Showing 2 changed files with 7 additions and 3 deletions.
4 changes: 3 additions & 1 deletion requirements-tpu.txt
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,8 @@ ray[default]
--find-links https://storage.googleapis.com/jax-releases/jaxlib_nightly_releases.html
torch==2.6.0.dev20241126+cpu
torchvision==0.20.0.dev20241126+cpu
torch_xla[tpu] @ https://storage.googleapis.com/pytorch-xla-releases/wheels/tpuvm/torch_xla-2.6.0.dev20241126-cp310-cp310-linux_x86_64.whl
torch_xla[tpu] @ https://storage.googleapis.com/pytorch-xla-releases/wheels/tpuvm/torch_xla-2.6.0.dev20241126-cp39-cp39-linux_x86_64.whl ; python_version == "3.9"
torch_xla[tpu] @ https://storage.googleapis.com/pytorch-xla-releases/wheels/tpuvm/torch_xla-2.6.0.dev20241126-cp310-cp310-linux_x86_64.whl ; python_version == "3.10"
torch_xla[tpu] @ https://storage.googleapis.com/pytorch-xla-releases/wheels/tpuvm/torch_xla-2.6.0.dev20241126-cp311-cp311-linux_x86_64.whl ; python_version == "3.11"
jaxlib==0.4.36.dev20241122
jax==0.4.36.dev20241122
6 changes: 4 additions & 2 deletions vllm/worker/tpu_model_runner.py
Original file line number Diff line number Diff line change
Expand Up @@ -126,8 +126,10 @@ def __init__(
logger.warning(
"The max_model_len (%d) is too large. This may degrade the "
"performance due to the insufficient smem size. Consider "
"setting --max-model-len to a smaller value.",
self.model_config.max_model_len)
"setting --max-model-len to a smaller value, like %d.",
self.model_config.max_model_len,
self.model_config.max_model_len /
(block_table_size / smem_size))

def load_model(self) -> None:
self.device = self.device_config.device
Expand Down

0 comments on commit bf0d97d

Please sign in to comment.