The fine-tuning resume path in scripts/train_voxcpm_finetune.py called
torch.load() without weights_only=True for the LoRA checkpoint, full-model
checkpoint, optimizer, and scheduler. Loading an attacker-supplied checkpoint
directory would therefore execute arbitrary code during unpickling.
Every inference-time loader (model/voxcpm.py, model/voxcpm2.py, LoRA loading)
already passes weights_only=True, and the project ships a test asserting LoRA
loading rejects malicious pickle payloads. This closes the remaining gap so the
resume path matches that posture.
Also:
- app.py: add a --host flag so the Gradio server can bind to 127.0.0.1 instead
of being hardwired to 0.0.0.0 (default unchanged to preserve current behavior).
The UI exposes an unauthenticated api_name="generate" endpoint.
- tests/test_torch_load_safety.py: AST guard asserting every torch.load across
src/, scripts/, app.py and lora_ft_webui.py sets weights_only=True, plus a
behavioral check that a malicious pickle is blocked.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Exposed 'seed' parameter in VoxCPMModel and VoxCPM2Model generation methods.
- Added PyTorch RNG seed setting before inference runs.
- Handled 'retry_badcase' seed adjustment by incrementing the seed value on retries.
- Exposed 'self.last_successful_seed' as a model attribute for UI integrations.
- Propagated 'seed' parameter to high-level pipeline class and CLI tools (cli.py).
- Added '--seed' flag to full-finetune and LoRA inference scripts.
- Configured validation audio generation in training script to use a fixed seed for objective comparison on TensorBoard.
- Added comprehensive unit tests in CLI test files to validate seed parsing and propagation.
- Updated English and Chinese READMEs with seed usage examples.
Pass manifest path via --manifest flag (required) instead of as a
positional argument, so the test exercises cmd_validate rather than
argparse error handling. Also assert returncode==1 and check stderr
for the FAILED/error message to prevent false positives.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Invalid audio rows (bad path or sample-rate mismatch) no longer
increment valid_samples; has_error is now set on any audio failure
- _check_audio_file now enforces the expected sample rate when soundfile
is available, making --sample-rate actually useful
- ref_audio missing-file warning is emitted for every invalid entry
independently, not only before the first valid one is seen
- New tests cover each of the four corrected behaviours: invalid audio
count, sample-rate mismatch, mixed ref_audio, and CLI exit code
LoRA is a first-class workflow in VoxCPM, and the project already prefers
safetensors plus weights-only fallback loading for base model artifacts. The
legacy LoRA .ckpt/.pth path was the remaining place that still deserialized
arbitrary pickle objects, so this switches it to weights_only=True and adds
focused regression coverage for both model loaders.
Constraint: Must preserve compatibility with tensor-only legacy LoRA checkpoints
Rejected: Remove .ckpt/.pth support entirely | too disruptive for existing users
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep LoRA artifact handling aligned with the existing safetensors-first, weights-only loading pattern
Tested: python3 -m pytest -q tests/test_lora_checkpoint_loading.py tests/test_model_utils.py -q
Not-tested: Full end-to-end LoRA hot-load with heavyweight model assets
Add a new `validate` subcommand that checks JSONL training manifests
before starting expensive fine-tuning jobs. This catches format issues,
missing audio files, and data quality problems early.
The validator performs:
- JSONL format validation (each line must be valid JSON)
- Required column checks (text, audio)
- Audio file existence and readability verification
- Duration and text length statistics (min, max, mean, median)
- Optional ref_audio column validation
- Warnings for very short (<0.3s) or very long (>30s) audio samples
Usage:
voxcpm validate --manifest train.jsonl
voxcpm validate --manifest train.jsonl --sample-rate 16000 --verbose
The module uses lazy imports for soundfile, so it works even in
minimal environments. Includes 11 unit tests covering all validation
paths.
Support optional ref_audio samples in finetuning and make runtime device selection explicit while keeping auto fallback behavior consistent. Also ignore the local app override file to avoid accidental commits.
Made-with: Cursor
ZGY