Labyricorn/voicebox
1
0
Fork 0
mirror of https://github.com/jamiepine/voicebox.git synced 2026-09-29 07:05:14 -07:00
Code Issues Packages Projects Releases Wiki Activity

fix(offline): remove process-global offline guard from Qwen3 LLM loads (#924)

Browse Source 
force_offline_if_cached flips HF_HUB_OFFLINE (env + huggingface_hub
constant + transformers._is_offline_mode) process-wide for the duration
of a cached LLM load, silently switching every concurrent model
download/load on other threads to offline mode. With default capture
settings (whisper-turbo STT + Qwen3 refinement + auto_refine) a first
run downloads several models concurrently, and a poisoned fetch
surfaces as "Can't load feature extractor..." (whisper) or
"Unrecognized model ... model_type" (Qwen3) rather than anything
mentioning offline mode.

These are the last two call sites of the guard — the same pattern was
deliberately removed app-wide in #524/#530 after identical failures,
and the 0.5.0 LLM backend reintroduced it. LLM loads now run with the
process's default HF_HUB_OFFLINE state, matching every other backend
(issue #462 precedent).

Fixes #841

Claude-Session: https://claude.ai/code/session_011iwL9AyeAWgz2jpgcHxJpC

Co-authored-by: Claude Fable 5 <[email protected]>
This commit is contained in:
Daniel Knoodle
2026-07-20 18:53:32 -07:00
committed by Jamie Pine
co-authored by Claude Fable 5
parent e001439c06
commit 3f4631c865
1 changed files with 15 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
backend/backends/qwen_llm_backend.py
+15 -12
View File
@@ -20,7 +20,6 @@ from .base import (
model_load_progress, model_load_progress,
) )
from ..services.mlx_thread import run_on_mlx_thread, clear_mlx_cache from ..services.mlx_thread import run_on_mlx_thread, clear_mlx_cache
from ..utils.hf_offline_patch import force_offline_if_cached
logger = logging.getLogger(__name__) logger = logging.getLogger(__name__)
@@ -104,15 +103,19 @@ class PyTorchQwenLLMBackend:
with model_load_progress(progress_model_name, is_cached): with model_load_progress(progress_model_name, is_cached):
logger.info("Loading Qwen3 %s on %s...", model_size, self.device) logger.info("Loading Qwen3 %s on %s...", model_size, self.device)
with force_offline_if_cached(is_cached, progress_model_name): # Loads run with the process's default HF_HUB_OFFLINE state.
self.tokenizer = AutoTokenizer.from_pretrained(repo) # Forcing offline for cached models flips process-global state
dtype = torch.float16 if self.device in ("cuda", "mps") else torch.float32 # and silently switches every concurrent download/load on other
self.model = AutoModelForCausalLM.from_pretrained( # threads to offline mode (issue #841) — the same regression
repo, # removed app-wide in #524/#530.
dtype=dtype, self.tokenizer = AutoTokenizer.from_pretrained(repo)
) dtype = torch.float16 if self.device in ("cuda", "mps") else torch.float32
self.model.to(self.device) self.model = AutoModelForCausalLM.from_pretrained(
self.model.eval() repo,
dtype=dtype,
)
self.model.to(self.device)
self.model.eval()
self._current_model_size = model_size self._current_model_size = model_size
self.model_size = model_size self.model_size = model_size
@@ -235,8 +238,8 @@ class MLXQwenLLMBackend:
with model_load_progress(progress_model_name, is_cached): with model_load_progress(progress_model_name, is_cached):
logger.info("Loading Qwen3 %s via MLX...", model_size) logger.info("Loading Qwen3 %s via MLX...", model_size)
with force_offline_if_cached(is_cached, progress_model_name): # See the PyTorch loader comment — no offline forcing (issue #841).
loaded = mlx_load(repo) loaded = mlx_load(repo)
# mlx_lm.load returns (model, tokenizer) by default and # mlx_lm.load returns (model, tokenizer) by default and
# (model, tokenizer, config) when return_config=True. # (model, tokenizer, config) when return_config=True.