Skip to content

Fix local GGUF settings reload falling through to transformers path #5332

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

Closed
BlackBox-Labs wants to merge 7 commits into from
Closed

Fix local GGUF settings reload falling through to transformers path #5332

Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
8a84475
Fix local GGUF settings reload falling through to transformers path
BlackBox-Labs May 8, 2026
a02dfd4
fix: narrow local GGUF reload condition with model_identifier match
BlackBox-Labs May 8, 2026
fcaadd9
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] May 8, 2026
7c0923a
fix: skip already_loaded for local GGUF in orchestrator else block
BlackBox-Labs May 8, 2026
977f406
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] May 8, 2026
d8e2e8f
Merge branch 'main' into fix/local-gguf-reload-settings
Imagineer99 May 8, 2026
58f8aa9
Merge branch 'main' into fix/local-gguf-reload-settings
rolandtannous May 14, 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
19 changes: 19 additions & 0 deletions studio/backend/routes/inference.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -517,6 +517,7 @@ async def load_model(
else:
if (
backend.active_model_name
and not backend.active_model_name.lower().endswith(".gguf")
and backend.active_model_name.lower() == model_identifier.lower()
):
logger.info(
Expand Down Expand Up @@ -573,6 +574,24 @@ async def load_model(
chat_template = _chat_template,
)

# ── Local GGUF already-loaded: kill before reload ──────────
# When settings change triggers a reload of a local GGUF model
# (no gguf_variant), kill the existing llama-server so the GGUF
# reload path below can start fresh with new params. Without this,
# the reload falls through to the transformers path, which fails
# because GGUF files don't have a HuggingFace config.json.
if (
not request.gguf_variant
and llama_backend.is_loaded
and llama_backend.model_identifier
and llama_backend.model_identifier.lower() == model_identifier.lower()
):
Comment on lines +584 to +588

@chatgpt-codex-connector chatgpt-codex-connector Bot May 8, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

P2 Badge Defer local GGUF unload until validations pass

This preemptive llama_backend.unload_model() runs before request validation, so a bad reload request can terminate a healthy serving model and then return an error. For example, when the same local GGUF is loaded and the caller sends gpu_ids, this branch unloads first, then the GGUF path rejects the request with 400 (gpu_ids is not supported for GGUF models yet), leaving inference with no model loaded. Please move this unload until after request/config validation that can fail.

Useful? React with 👍 / 👎.

logger.info(
f"Local GGUF already loaded, reloading with updated settings: "
f"{model_log_label}"
)
llama_backend.unload_model()

# Create config using clean factory method
# is_lora is auto-detected from adapter_config.json on disk/HF
config = ModelConfig.from_identifier(
Expand Down