Skip to content

fix: accept 0-indexed safetensors shard names in CI weight validator #24237

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
Kangyan-Zhou merged 2 commits into from
May 2, 2026
+4 −1
Merged

fix: accept 0-indexed safetensors shard names in CI weight validator #24237

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9c50bd0
fix: accept 0-indexed safetensors shard names in CI weight validator
alisonshao May 1, 2026
5d8263a
drop unit-test addition: test/manual/ is not run in CI
alisonshao May 1, 2026
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
5 changes: 4 additions & 1 deletion python/sglang/srt/model_loader/ci_weight_validation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1426,7 +1426,10 @@ def _validate_sharded_model(
for group_key, group_info in shard_groups.items():
total_shards = group_info["total"]
found_shards = set(group_info["found_shards"])
expected_shards = set(range(1, total_shards + 1))
# Shards may be 0-indexed (e.g. inclusionAI/Ring-2.5-1T) or 1-indexed
# (e.g. deepseek-ai/DeepSeek-V3); both are valid HF conventions.
min_idx = min(found_shards) if found_shards else 1
expected_shards = set(range(min_idx, min_idx + total_shards))
Comment on lines +1431 to +1432
Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist Bot May 1, 2026

Choose a reason for hiding this comment

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

medium

The logic using min(found_shards) to determine the starting index is fragile. If the first few shards are missing from the cache, min(found_shards) will shift the expected range, leading to confusing error messages (e.g., reporting shards beyond the actual total count as missing). Since Hugging Face models follow either 0-indexed or 1-indexed conventions, a more robust heuristic is to check if shard 0 is present. If it is, assume 0-indexing; otherwise, default to 1-indexing (which is the standard convention). Additionally, please note that validate_cache_lightweight (around line 1213) contains the same hardcoded 1-indexed logic and should be updated for consistency, although it is outside the current diff hunks.

Suggested change
min_idx = min(found_shards) if found_shards else 1
expected_shards = set(range(min_idx, min_idx + total_shards))
min_idx = 0 if 0 in found_shards else 1
expected_shards = set(range(min_idx, min_idx + total_shards))


# Check for missing shards
missing_shards = expected_shards - found_shards
Expand Down
Loading