Skip to content

fix:guard torch.compile in get_smol so first index doesnt crash in windows #30

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
Qayad-Ali wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix:guard torch.compile in get_smol so first index doesnt crash in windows #30

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
13 changes: 11 additions & 2 deletions src/gurrt/core/models.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,5 @@
import torch
import sys
from transformers import (
CLIPProcessor,
CLIPModel,
Expand Down Expand Up @@ -79,7 +80,15 @@ def get_smol(self, flag:bool):

self._smol = SmolVLMForConditionalGeneration.from_pretrained(path,
local_files_only= True)
self._smol = torch.compile(self._smol, mode="reduce-overhead")
## torch.compile works with Triton, which is not on Windows, so this one
# line crashes the very first index. Only run it when we're on a CUDA GPU
# and not on Windows — and if the compile still chokes, just skip it instead
# of taking whole run down with it.
if self.device == "cuda" and sys.platform != "win32":
try:
self._smol = torch.compile(self._smol, mode="reduce-overhead")
except Exception as e:
ui.warn(f"torch.compile unavailable -- continuing without it ({e})")
return self._to_device(self._smol), self._smol_processor

def release_smol(self):
Expand Down Expand Up @@ -149,4 +158,4 @@ def download_models(cache_dir):
snapshot_download(
repo_id="Systran/faster-distil-whisper-large-v2",
local_dir=str(cache_dir / "whisper_model"),
)
)