Bump version: 0.4.4 → 0.4.5

* fix(offline): patch transformers mistral-regex check to survive HF failures

transformers 4.57.x's `PreTrainedTokenizerBase._patch_mistral_regex` calls
`huggingface_hub.model_info(repo_id)` unconditionally during any non-local
tokenizer load to probe for Mistral-family models. The call raises on
`HF_HUB_OFFLINE=1`, on network outages, and on slow/blocked HF endpoints,
and transformers doesn't catch any of it — the exception bubbles out of
`from_pretrained` and kills the load for unrelated engines (Qwen TTS,
Qwen CustomVoice, TADA, etc.).

0.4.2's load-time `force_offline_if_cached` guard walked straight into
this trap: on cached online users it flipped `HF_HUB_OFFLINE=1` and
converted a healthy load into a hard crash. 0.4.3's inference-path guard
masked it; #524 removed the inference guard in 0.4.4, and users updating
to 0.4.4 started hitting the same error on the load path instead
(#526).

Fix:
- Wrap `_patch_mistral_regex` so any exception from the inner HF
  metadata check is swallowed and the tokenizer is returned unchanged.
  Voicebox never loads Mistral models, so the regex rewrite this check
  gates is a no-op for us; matches the success-path behavior for
  non-Mistral repos (tokenization_utils_base.py:2503).
- Drop the `force_offline_if_cached` wraps from every load path
  (pytorch_backend Qwen + Whisper, qwen_custom_voice_backend,
  mlx_backend Qwen + Whisper). With the mistral patch in place they
  provide zero value and only risk re-introducing the same class of
  bug. Helper and its unit tests stay — still correct for targeted
  future use.
- Add `backend/tests/test_offline_patch.py` covering
  OfflineModeIsEnabled / ConnectionError suppression, success
  pass-through, idempotence, and the missing-method no-op path.

Fixes #526.

* fix(offline): install mistral-regex patch for non-MLX backends

The previous commit left the patch wired only through ``mlx_backend.py``'s
existing import of ``hf_offline_patch``. On Windows/Linux/CUDA users who
never load the MLX backend (everyone who hit #526), the patch module was
never imported, so ``patch_transformers_mistral_regex`` never ran and the
crash persisted.

Hoist the import into ``backends/__init__.py``. Every backend imports from
this package, so the module-level patch install runs before any
``from_pretrained`` call regardless of which engine the user picks.

Caught by CodeRabbit and Cursor Bugbot on #530.

.bumpversion.cfg

@@ -1,5 +1,5 @@
 [bumpversion]
-current_version = 0.4.3
+current_version = 0.4.5
 commit = True
 tag = True
 tag_name = v{new_version}

CHANGELOG.md

@@ -7,6 +7,25 @@
 
 ## [Unreleased]
 
+## [0.4.5] - 2026-04-22
+
+Second hotfix for the "offline mode is enabled" crash on model load. 0.4.4 reverted the inference-path offline guards but kept the same trap on the load path, so users who updated to 0.4.4 kept hitting the exact error the release was supposed to fix ([#526](https://github.com/jamiepine/voicebox/issues/526)). This release removes the load-path guards and patches the transformers tokenizer load to be robust to HuggingFace metadata failures at the source, so the class of bug can't recur.
+
+### Reliability
+
+- **Load no longer fails with "offline mode is enabled"** ([#530](https://github.com/jamiepine/voicebox/pull/530), fixes [#526](https://github.com/jamiepine/voicebox/issues/526)). transformers 4.57.x added an unconditional `huggingface_hub.model_info()` call inside `AutoTokenizer.from_pretrained` (via `_patch_mistral_regex`) that runs for every non-local repo load, regardless of cache state or whether the target model is actually a Mistral variant. The load-time `HF_HUB_OFFLINE` guard from 0.4.2 turned that into a hard crash for cached online users the moment 0.4.4 removed the inference-path guard that had been masking the problem. Fix wraps `_patch_mistral_regex` so any exception from the HF metadata check is caught and the tokenizer is returned unchanged — matching the success-path behavior for non-Mistral repos. The wrapper installs at `backend.backends` import time so it covers Qwen Base, Qwen CustomVoice, TADA, and every other transformers-backed engine on Windows, Linux, and CUDA alike. The load-time `force_offline_if_cached` guards were removed — with the wrapper in place they provide zero value and only risk re-introducing the same failure mode.
+- **No more 30s pause when generating without a network.** The HuggingFace metadata timeout called out as a known caveat in 0.4.4 is covered by the same patch; offline users no longer wait for the check to time out before load completes.
+
+## [0.4.4] - 2026-04-21
+
+Hotfix for a regression in 0.4.3 where generation and transcription could fail outright with "offline mode is enabled" even when the user was online.
+
+### Reliability
+
+- **Inference no longer fails with "offline mode is enabled" while online** ([#524](https://github.com/jamiepine/voicebox/pull/524), reverts the inference-path guards from [#503](https://github.com/jamiepine/voicebox/pull/503)). 0.4.3 wrapped every inference body (`generate`, `transcribe`, `create_voice_clone_prompt`) with a process-wide `HF_HUB_OFFLINE` flip to stop lazy HuggingFace lookups from hanging when the network drops mid-inference ([#462](https://github.com/jamiepine/voicebox/issues/462)). That flag also blocks legitimate metadata calls (e.g. `HfApi().model_info` for revision resolution) so online users started seeing generation fail outright. Inference now runs with the process's default HF state. Load-time offline guards — which weren't the source of the regression — stay in place.
+
+**Known caveat**: users generating without an internet connection may see brief pauses during inference while HuggingFace metadata lookups time out (typically ~30s, after which the library recovers). A proper offline-mode toggle is planned for 0.4.5.
+
 ## [0.4.3] - 2026-04-20
 
 A patch focused on two user-impacting reliability fixes: macOS DMG notarization (unblocks `brew install voicebox` on macOS 15 Sequoia and fixes spurious "app isn't signed" Gatekeeper dialogs on older Intel Macs) and Kokoro Japanese voice initialization on fresh installs.
@@ -638,7 +657,9 @@ The first public release of Voicebox — an open-source voice synthesis studio p
 
 Tauri v2, React, TypeScript, Tailwind CSS, FastAPI, Qwen3-TTS, Whisper, SQLite
 
-[Unreleased]: https://github.com/jamiepine/voicebox/compare/v0.4.3...HEAD
+[Unreleased]: https://github.com/jamiepine/voicebox/compare/v0.4.5...HEAD
+[0.4.5]: https://github.com/jamiepine/voicebox/compare/v0.4.4...v0.4.5
+[0.4.4]: https://github.com/jamiepine/voicebox/compare/v0.4.3...v0.4.4
 [0.4.3]: https://github.com/jamiepine/voicebox/compare/v0.4.2...v0.4.3
 [0.4.2]: https://github.com/jamiepine/voicebox/compare/v0.4.1...v0.4.2
 [0.4.1]: https://github.com/jamiepine/voicebox/compare/v0.4.0...v0.4.1

app/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@voicebox/app",
-  "version": "0.4.3",
+  "version": "0.4.5",
   "private": true,
   "type": "module",
   "scripts": {

backend/__init__.py

@@ -1,3 +1,3 @@
 # Backend package
 
-__version__ = "0.4.3"
+__version__ = "0.4.5"

backend/backends/__init__.py

@@ -5,6 +5,13 @@ Provides a unified interface for MLX and PyTorch backends,
 and a model config registry that eliminates per-engine dispatch maps.
 """
 
+# Install HF compatibility patches before any backend imports transformers /
+# huggingface_hub. The module runs ``patch_transformers_mistral_regex`` at
+# import time, which wraps transformers' tokenizer load against the
+# unconditional HuggingFace metadata call that otherwise raises on
+# HF_HUB_OFFLINE=1 and on network failures.
+from ..utils import hf_offline_patch  # noqa: F401
+
 import threading
 from dataclasses import dataclass, field
 from typing import Protocol, Optional, Tuple, List

backend/backends/mlx_backend.py

@@ -20,7 +20,6 @@ ensure_original_qwen_config_cached()
 from . import TTSBackend, STTBackend, LANGUAGE_CODE_TO_NAME, WHISPER_HF_REPOS
 from .base import is_model_cached, combine_voice_prompts as _combine_voice_prompts, model_load_progress
 from ..utils.cache import get_cache_key, get_cached_voice_prompt, cache_voice_prompt
-from ..utils.hf_offline_patch import force_offline_if_cached
 
 
 class MLXTTSBackend:
@@ -99,8 +98,7 @@ class MLXTTSBackend:
 
             logger.info("Loading MLX TTS model %s...", model_size)
 
-            with force_offline_if_cached(is_cached, model_name):
-                self.model = load(model_path)
+            self.model = load(model_path)
 
         self._current_model_size = model_size
         self.model_size = model_size
@@ -193,8 +191,6 @@ class MLXTTSBackend:
 
         logger.info("Generating audio for text: %s", text)
 
-        model_name = f"qwen-tts-{self._current_model_size}"
-
         def _generate_sync():
             """Run synchronous generation in thread pool."""
             # MLX generate() returns a generator yielding GenerationResult objects
@@ -220,40 +216,38 @@ class MLXTTSBackend:
                 logger.warning("Regenerating without voice prompt.")
                 ref_audio = None
 
-            # Model is loaded → weights are on disk. Force offline so
-            # lazy tokenizer/config lookups inside mlx_audio don't hang
-            # when the user is disconnected (issue #462).
-            with force_offline_if_cached(True, model_name):
-                # Check if model supports voice cloning via generate method
-                # MLX API may support ref_audio parameter directly
-                try:
-                    # Try with voice cloning parameters if supported
-                    if ref_audio:
-                        # Check if generate accepts ref_audio parameter
-                        import inspect
+            # Inference runs with the process's default HF_HUB_OFFLINE
+            # state. Forcing offline here (previously used to avoid lazy
+            # mlx_audio lookups hanging when the network drops mid-inference,
+            # issue #462) regressed online users because libraries make
+            # legitimate metadata calls during generation.
+            try:
+                if ref_audio:
+                    # Check if generate accepts ref_audio parameter
+                    import inspect
 
-                        sig = inspect.signature(self.model.generate)
-                        if "ref_audio" in sig.parameters:
-                            # Generate with voice cloning
-                            for result in self.model.generate(text, ref_audio=ref_audio, ref_text=ref_text, lang_code=lang):
-                                audio_chunks.append(np.array(result.audio))
-                                sample_rate = result.sample_rate
-                        else:
-                            # Fallback: generate without voice cloning
-                            for result in self.model.generate(text, lang_code=lang):
-                                audio_chunks.append(np.array(result.audio))
-                                sample_rate = result.sample_rate
+                    sig = inspect.signature(self.model.generate)
+                    if "ref_audio" in sig.parameters:
+                        # Generate with voice cloning
+                        for result in self.model.generate(text, ref_audio=ref_audio, ref_text=ref_text, lang_code=lang):
+                            audio_chunks.append(np.array(result.audio))
+                            sample_rate = result.sample_rate
                     else:
-                        # No voice prompt, generate normally
+                        # Fallback: generate without voice cloning
                         for result in self.model.generate(text, lang_code=lang):
                             audio_chunks.append(np.array(result.audio))
                             sample_rate = result.sample_rate
-                except Exception as e:
-                    # If voice cloning fails, try without it
-                    logger.warning("Voice cloning failed, generating without voice prompt: %s", e)
+                else:
+                    # No voice prompt, generate normally
                     for result in self.model.generate(text, lang_code=lang):
                         audio_chunks.append(np.array(result.audio))
                         sample_rate = result.sample_rate
+            except Exception as e:
+                # If voice cloning fails, try without it
+                logger.warning("Voice cloning failed, generating without voice prompt: %s", e)
+                for result in self.model.generate(text, lang_code=lang):
+                    audio_chunks.append(np.array(result.audio))
+                    sample_rate = result.sample_rate
 
             # Concatenate all chunks
             if audio_chunks:
@@ -315,8 +309,7 @@ class MLXSTTBackend:
             model_name = WHISPER_HF_REPOS.get(model_size, f"openai/whisper-{model_size}")
             logger.info("Loading MLX Whisper model %s...", model_size)
 
-            with force_offline_if_cached(is_cached, progress_model_name):
-                self.model = load(model_name)
+            self.model = load(model_name)
 
         self.model_size = model_size
         logger.info("MLX Whisper model %s loaded successfully", model_size)
@@ -347,8 +340,6 @@ class MLXSTTBackend:
         """
         await self.load_model_async(model_size)
 
-        progress_model_name = f"whisper-{self.model_size}"
-
         def _transcribe_sync():
             """Run synchronous transcription in thread pool."""
             # MLX Whisper transcription using generate method
@@ -357,11 +348,10 @@ class MLXSTTBackend:
             if language:
                 decode_options["language"] = language
 
-            # Model is loaded → weights are on disk. Force offline so
-            # lazy tokenizer/config lookups don't hang when the user is
-            # disconnected (issue #462).
-            with force_offline_if_cached(True, progress_model_name):
-                result = self.model.generate(str(audio_path), **decode_options)
+            # Inference runs with the process's default HF_HUB_OFFLINE
+            # state — see the comment in MLXTTSBackend.generate for the
+            # regression this revert fixes (issue #462).
+            result = self.model.generate(str(audio_path), **decode_options)
 
             # Extract text from result
             if isinstance(result, str):

backend/backends/pytorch_backend.py

@@ -21,7 +21,6 @@ from .base import (
 )
 from ..utils.cache import get_cache_key, get_cached_voice_prompt, cache_voice_prompt
 from ..utils.audio import load_audio
-from ..utils.hf_offline_patch import force_offline_if_cached
 
 
 class PyTorchTTSBackend:
@@ -106,21 +105,20 @@ class PyTorchTTSBackend:
             from huggingface_hub import constants as hf_constants
             tts_cache_dir = hf_constants.HF_HUB_CACHE
 
-            with force_offline_if_cached(is_cached, model_name):
-                if self.device == "cpu":
-                    self.model = Qwen3TTSModel.from_pretrained(
-                        model_path,
-                        cache_dir=tts_cache_dir,
-                        torch_dtype=torch.float32,
-                        low_cpu_mem_usage=False,
-                    )
-                else:
-                    self.model = Qwen3TTSModel.from_pretrained(
-                        model_path,
-                        cache_dir=tts_cache_dir,
-                        device_map=self.device,
-                        torch_dtype=torch.bfloat16,
-                    )
+            if self.device == "cpu":
+                self.model = Qwen3TTSModel.from_pretrained(
+                    model_path,
+                    cache_dir=tts_cache_dir,
+                    torch_dtype=torch.float32,
+                    low_cpu_mem_usage=False,
+                )
+            else:
+                self.model = Qwen3TTSModel.from_pretrained(
+                    model_path,
+                    cache_dir=tts_cache_dir,
+                    device_map=self.device,
+                    torch_dtype=torch.bfloat16,
+                )
 
         self._current_model_size = model_size
         self.model_size = model_size
@@ -172,19 +170,17 @@ class PyTorchTTSBackend:
                     # This shouldn't happen in practice, but handle it
                     return {"prompt": cached_prompt}, True
 
-        model_name = f"qwen-tts-{self._current_model_size}"
-
         def _create_prompt_sync():
             """Run synchronous voice prompt creation in thread pool."""
-            # Model is loaded → weights are on disk. Force offline so
-            # lazy tokenizer/config lookups inside qwen_tts don't hang
-            # when the user is disconnected (issue #462).
-            with force_offline_if_cached(True, model_name):
-                return self.model.create_voice_clone_prompt(
-                    ref_audio=str(audio_path),
-                    ref_text=reference_text,
-                    x_vector_only_mode=False,
-                )
+            # Inference runs with the process's default HF_HUB_OFFLINE
+            # state. Forcing offline here (issue #462) regressed online
+            # users whose libraries issue legitimate metadata lookups
+            # during voice-prompt creation.
+            return self.model.create_voice_clone_prompt(
+                ref_audio=str(audio_path),
+                ref_text=reference_text,
+                x_vector_only_mode=False,
+            )
 
         # Run blocking operation in thread pool
         voice_prompt_items = await asyncio.to_thread(_create_prompt_sync)
@@ -227,24 +223,20 @@ class PyTorchTTSBackend:
         # Load model
         await self.load_model_async(None)
 
-        model_name = f"qwen-tts-{self._current_model_size}"
-
         def _generate_sync():
             """Run synchronous generation in thread pool."""
             # Set seed if provided
             if seed is not None:
                 manual_seed(seed, self.device)
 
-            # Model is loaded → weights are on disk. Force offline so
-            # lazy tokenizer/config lookups inside qwen_tts don't hang
-            # when the user is disconnected (issue #462).
-            with force_offline_if_cached(True, model_name):
-                wavs, sample_rate = self.model.generate_voice_clone(
-                    text=text,
-                    voice_clone_prompt=voice_prompt,
-                    language=LANGUAGE_CODE_TO_NAME.get(language, "auto"),
-                    instruct=instruct,
-                )
+            # See _create_prompt_sync comment — inference runs with the
+            # process's default HF_HUB_OFFLINE state (issue #462).
+            wavs, sample_rate = self.model.generate_voice_clone(
+                text=text,
+                voice_clone_prompt=voice_prompt,
+                language=LANGUAGE_CODE_TO_NAME.get(language, "auto"),
+                instruct=instruct,
+            )
             return wavs[0], sample_rate
 
         # Run blocking inference in thread pool to avoid blocking event loop
@@ -303,9 +295,8 @@ class PyTorchSTTBackend:
             model_name = WHISPER_HF_REPOS.get(model_size, f"openai/whisper-{model_size}")
             logger.info("Loading Whisper model %s on %s...", model_size, self.device)
 
-            with force_offline_if_cached(is_cached, progress_model_name):
-                self.processor = WhisperProcessor.from_pretrained(model_name)
-                self.model = WhisperForConditionalGeneration.from_pretrained(model_name)
+            self.processor = WhisperProcessor.from_pretrained(model_name)
+            self.model = WhisperForConditionalGeneration.from_pretrained(model_name)
 
         self.model.to(self.device)
         self.model_size = model_size
@@ -342,46 +333,44 @@ class PyTorchSTTBackend:
         """
         await self.load_model_async(model_size)
 
-        progress_model_name = f"whisper-{self.model_size}"
-
         def _transcribe_sync():
             """Run synchronous transcription in thread pool."""
             # Load audio
             audio, _sr = load_audio(audio_path, sample_rate=16000)
 
-            # Model is loaded → weights are on disk. Force offline so
-            # `get_decoder_prompt_ids` and any lazy tokenizer lookups
-            # don't hang when the user is disconnected (issue #462).
-            with force_offline_if_cached(True, progress_model_name):
-                # Process audio
-                inputs = self.processor(
-                    audio,
-                    sampling_rate=16000,
-                    return_tensors="pt",
+            # Inference runs with the process's default HF_HUB_OFFLINE
+            # state — forcing offline here (issue #462) broke online users
+            # whose `get_decoder_prompt_ids` / tokenizer calls issue
+            # legitimate metadata lookups.
+            # Process audio
+            inputs = self.processor(
+                audio,
+                sampling_rate=16000,
+                return_tensors="pt",
+            )
+            inputs = inputs.to(self.device)
+
+            # Generate transcription
+            # If language is provided, force it; otherwise let Whisper auto-detect
+            generate_kwargs = {}
+            if language:
+                forced_decoder_ids = self.processor.get_decoder_prompt_ids(
+                    language=language,
+                    task="transcribe",
                 )
-                inputs = inputs.to(self.device)
+                generate_kwargs["forced_decoder_ids"] = forced_decoder_ids
 
-                # Generate transcription
-                # If language is provided, force it; otherwise let Whisper auto-detect
-                generate_kwargs = {}
-                if language:
-                    forced_decoder_ids = self.processor.get_decoder_prompt_ids(
-                        language=language,
-                        task="transcribe",
-                    )
-                    generate_kwargs["forced_decoder_ids"] = forced_decoder_ids
+            with torch.no_grad():
+                predicted_ids = self.model.generate(
+                    inputs["input_features"],
+                    **generate_kwargs,
+                )
 
-                with torch.no_grad():
-                    predicted_ids = self.model.generate(
-                        inputs["input_features"],
-                        **generate_kwargs,
-                    )
-
-                # Decode
-                transcription = self.processor.batch_decode(
-                    predicted_ids,
-                    skip_special_tokens=True,
-                )[0]
+            # Decode
+            transcription = self.processor.batch_decode(
+                predicted_ids,
+                skip_special_tokens=True,
+            )[0]
 
             return transcription.strip()
 

backend/backends/qwen_custom_voice_backend.py

@@ -28,7 +28,6 @@ from .base import (
     combine_voice_prompts as _combine_voice_prompts,
     model_load_progress,
 )
-from ..utils.hf_offline_patch import force_offline_if_cached
 
 logger = logging.getLogger(__name__)
 
@@ -105,19 +104,18 @@ class QwenCustomVoiceBackend:
             model_path = self._get_model_path(model_size)
             logger.info("Loading Qwen CustomVoice %s on %s...", model_size, self.device)
 
-            with force_offline_if_cached(is_cached, model_name):
-                if self.device == "cpu":
-                    self.model = Qwen3TTSModel.from_pretrained(
-                        model_path,
-                        torch_dtype=torch.float32,
-                        low_cpu_mem_usage=False,
-                    )
-                else:
-                    self.model = Qwen3TTSModel.from_pretrained(
-                        model_path,
-                        device_map=self.device,
-                        torch_dtype=torch.bfloat16,
-                    )
+            if self.device == "cpu":
+                self.model = Qwen3TTSModel.from_pretrained(
+                    model_path,
+                    torch_dtype=torch.float32,
+                    low_cpu_mem_usage=False,
+                )
+            else:
+                self.model = Qwen3TTSModel.from_pretrained(
+                    model_path,
+                    device_map=self.device,
+                    torch_dtype=torch.bfloat16,
+                )
 
         self._current_model_size = model_size
         self.model_size = model_size
@@ -186,7 +184,6 @@ class QwenCustomVoiceBackend:
         await self.load_model_async(None)
 
         speaker = voice_prompt.get("preset_voice_id") or QWEN_CV_DEFAULT_SPEAKER
-        model_name = f"qwen-custom-voice-{self._current_model_size}"
 
         def _generate_sync():
             if seed is not None:
@@ -206,11 +203,11 @@ class QwenCustomVoiceBackend:
             if instruct:
                 kwargs["instruct"] = instruct
 
-            # Model is loaded → weights are on disk. Force offline so
-            # lazy tokenizer/config lookups inside qwen_tts don't hang
-            # when the user is disconnected (issue #462).
-            with force_offline_if_cached(True, model_name):
-                wavs, sample_rate = self.model.generate_custom_voice(**kwargs)
+            # Inference runs with the process's default HF_HUB_OFFLINE
+            # state. Forcing offline here (issue #462) regressed online
+            # users whose libraries issue legitimate metadata lookups
+            # during generation.
+            wavs, sample_rate = self.model.generate_custom_voice(**kwargs)
             return wavs[0], sample_rate
 
         audio, sample_rate = await asyncio.to_thread(_generate_sync)

backend/tests/test_offline_patch.py

@@ -0,0 +1,113 @@
+"""
+Unit tests for ``patch_transformers_mistral_regex``.
+
+Verifies that our wrapper around
+``transformers.PreTrainedTokenizerBase._patch_mistral_regex`` catches
+exceptions from the unconditional ``huggingface_hub.model_info()`` lookup
+and returns the tokenizer unchanged — matching the success-path behavior
+for non-Mistral repos (transformers 4.57.3, ``tokenization_utils_base.py:2503``).
+
+NOTE: These tests mutate ``transformers.PreTrainedTokenizerBase`` globally;
+run serially, not under ``pytest-xdist`` with per-worker process isolation.
+"""
+
+import sys
+from pathlib import Path
+
+import pytest
+
+sys.path.insert(0, str(Path(__file__).parent.parent))
+
+from huggingface_hub.errors import OfflineModeIsEnabled  # noqa: E402
+from transformers.tokenization_utils_base import PreTrainedTokenizerBase  # noqa: E402
+
+import utils.hf_offline_patch as hf_offline_patch  # noqa: E402
+
+
+@pytest.fixture(autouse=True)
+def restore_mistral_regex():
+    """Snapshot the current ``_patch_mistral_regex`` and restore after each test."""
+    saved = PreTrainedTokenizerBase.__dict__.get("_patch_mistral_regex")
+    saved_flag = hf_offline_patch._mistral_regex_patched
+    try:
+        yield
+    finally:
+        if saved is not None:
+            PreTrainedTokenizerBase._patch_mistral_regex = saved
+        hf_offline_patch._mistral_regex_patched = saved_flag
+
+
+def _apply_patch():
+    hf_offline_patch._mistral_regex_patched = False
+    hf_offline_patch.patch_transformers_mistral_regex()
+
+
+def test_suppresses_offline_mode_is_enabled(monkeypatch):
+    _apply_patch()
+
+    import huggingface_hub
+
+    def raise_offline(*_args, **_kwargs):
+        raise OfflineModeIsEnabled("offline")
+
+    monkeypatch.setattr(huggingface_hub, "model_info", raise_offline)
+
+    sentinel = object()
+    result = PreTrainedTokenizerBase._patch_mistral_regex(
+        sentinel, "Qwen/Qwen3-TTS-12Hz-1.7B-Base"
+    )
+    assert result is sentinel
+
+
+def test_suppresses_connection_errors(monkeypatch):
+    _apply_patch()
+
+    import huggingface_hub
+
+    def raise_connection(*_args, **_kwargs):
+        raise ConnectionError("network unreachable")
+
+    monkeypatch.setattr(huggingface_hub, "model_info", raise_connection)
+
+    sentinel = object()
+    result = PreTrainedTokenizerBase._patch_mistral_regex(
+        sentinel, "Qwen/Qwen3-TTS-12Hz-1.7B-Base"
+    )
+    assert result is sentinel
+
+
+def test_passthrough_on_success(monkeypatch):
+    """When model_info returns non-Mistral tags the original falls through and returns the tokenizer unchanged."""
+    _apply_patch()
+
+    import huggingface_hub
+
+    class FakeInfo:
+        tags = ["model-type:qwen", "language:en"]
+
+    monkeypatch.setattr(huggingface_hub, "model_info", lambda *_a, **_kw: FakeInfo())
+
+    sentinel = object()
+    result = PreTrainedTokenizerBase._patch_mistral_regex(
+        sentinel, "Qwen/Qwen3-TTS-12Hz-1.7B-Base"
+    )
+    assert result is sentinel
+
+
+def test_idempotent():
+    _apply_patch()
+    first = PreTrainedTokenizerBase._patch_mistral_regex
+    hf_offline_patch.patch_transformers_mistral_regex()
+    second = PreTrainedTokenizerBase._patch_mistral_regex
+    assert first.__func__ is second.__func__
+
+
+def test_missing_method_is_noop(monkeypatch):
+    monkeypatch.delattr(PreTrainedTokenizerBase, "_patch_mistral_regex", raising=False)
+    hf_offline_patch._mistral_regex_patched = False
+    hf_offline_patch.patch_transformers_mistral_regex()
+    assert hf_offline_patch._mistral_regex_patched is False
+
+
+if __name__ == "__main__":
+    pytest.main([__file__, "-v"])

backend/utils/hf_offline_patch.py

@@ -142,6 +142,57 @@ def force_offline_if_cached(is_cached: bool, model_label: str = ""):
                 _saved_transformers_const = None
 
 
+_mistral_regex_patched = False
+
+
+def patch_transformers_mistral_regex():
+    """Make transformers' tokenizer load robust to HuggingFace metadata failures.
+
+    transformers 4.57.x added ``PreTrainedTokenizerBase._patch_mistral_regex``
+    which unconditionally calls ``huggingface_hub.model_info(repo_id)`` during
+    every non-local tokenizer load to check whether the model is a Mistral
+    variant. That call raises on ``HF_HUB_OFFLINE=1`` and on plain network
+    failures, killing unrelated loads (Qwen TTS, TADA, etc.).
+
+    Voicebox never loads Mistral models, so the rewrite the function would
+    apply is a no-op for us anyway. Wrap the method so any exception from the
+    metadata lookup returns the tokenizer unchanged — matching the success-path
+    behavior for non-Mistral repos (transformers 4.57.3,
+    ``tokenization_utils_base.py:2503``).
+    """
+    global _mistral_regex_patched
+    if _mistral_regex_patched:
+        return
+
+    try:
+        from transformers.tokenization_utils_base import PreTrainedTokenizerBase
+    except ImportError:
+        logger.debug("transformers not available, skipping mistral-regex patch")
+        return
+
+    original = getattr(PreTrainedTokenizerBase, "_patch_mistral_regex", None)
+    if original is None:
+        logger.debug(
+            "transformers has no _patch_mistral_regex attribute, skipping patch",
+        )
+        return
+
+    def safe_patch_mistral_regex(cls, tokenizer, pretrained_model_name_or_path, *args, **kwargs):
+        try:
+            return original(tokenizer, pretrained_model_name_or_path, *args, **kwargs)
+        except Exception as exc:
+            logger.debug(
+                "[mistral-regex-patch] suppressed %s for %r, returning tokenizer as-is",
+                type(exc).__name__,
+                pretrained_model_name_or_path,
+            )
+            return tokenizer
+
+    PreTrainedTokenizerBase._patch_mistral_regex = classmethod(safe_patch_mistral_regex)
+    _mistral_regex_patched = True
+    logger.debug("installed _patch_mistral_regex wrapper")
+
+
 def patch_huggingface_hub_offline():
     """Monkey-patch huggingface_hub to force offline mode."""
     try:
@@ -215,4 +266,5 @@ def ensure_original_qwen_config_cached():
 
 if os.environ.get("VOICEBOX_OFFLINE_PATCH", "1") != "0":
     patch_huggingface_hub_offline()
+    patch_transformers_mistral_regex()
     ensure_original_qwen_config_cached()

landing/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@voicebox/landing",
-  "version": "0.4.3",
+  "version": "0.4.5",
   "description": "Landing page for voicebox.sh",
   "scripts": {
     "dev": "next dev --turbo",

package.json

@@ -1,6 +1,6 @@
 {
   "name": "voicebox",
-  "version": "0.4.3",
+  "version": "0.4.5",
   "private": true,
   "workspaces": [
     "app",

tauri/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@voicebox/tauri",
   "private": true,
-  "version": "0.4.3",
+  "version": "0.4.5",
   "type": "module",
   "scripts": {
     "dev": "vite",

tauri/src-tauri/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "voicebox"
-version = "0.4.3"
+version = "0.4.5"
 description = "A production-quality desktop app for Qwen3-TTS voice cloning and generation"
 authors = ["you"]
 license = ""

tauri/src-tauri/tauri.conf.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://schema.tauri.app/config/2",
   "productName": "Voicebox",
-  "version": "0.4.3",
+  "version": "0.4.5",
   "identifier": "sh.voicebox.app",
   "build": {
     "beforeDevCommand": "bun run dev",

web/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@voicebox/web",
   "private": true,
-  "version": "0.4.3",
+  "version": "0.4.5",
   "type": "module",
   "scripts": {
     "dev": "vite",