unslothai/unsloth · error · ValueError

No valid examples after SNAC preprocessing (skipped {skipped

Error message

No valid examples after SNAC preprocessing (skipped {skipped})

What it means

ValueError raised after the SNAC encoding loop when processed_examples is empty — every example failed during codec encoding and was skipped. The SNAC model is loaded, audio encoded per example, and failures counted; if all rows fail, the error reports the skipped count and aborts before Dataset.from_list. The per-example exceptions were logged as warnings during the run.

Source

Thrown at studio/backend/core/training/trainer.py:2021

            except Exception as e:
                logger.warning(f"Error processing SNAC example {idx}: {e}")
                skipped += 1
                continue

            if (idx + 1) % 100 == 0:
                self._update_progress(status_message = f"Encoding audio... {idx + 1}/{len(dataset)}")

        logger.info("Freeing SNAC codec model from GPU...\n")
        snac_model.to("cpu")
        del snac_model

        gc.collect()

        clear_gpu_cache()
        self._cuda_audio_used = True

        if not processed_examples:
            raise ValueError(f"No valid examples after SNAC preprocessing (skipped {skipped})")

        result_dataset = Dataset.from_list(processed_examples)
        logger.info(
            f"SNAC preprocessing complete: {len(result_dataset)} examples " f"({skipped} skipped)\n"
        )
        return result_dataset

    def _preprocess_bicodec_dataset(
        self,
        dataset,
        custom_format_mapping = None,
    ):
        """Preprocess dataset for Spark-TTS training with BiCodec tokenizer.

        Mirrors Spark_TTS_(0_5B).ipynb: encode audio with BiCodec (semantic +
        global tokens), format as special-token text strings for SFTTrainer
        with dataset_text_field="text".
        """

View on GitHub (pinned to 203007d190)

Solutions

Inspect the loop's logged per-example warnings to identify the recurring exception.
Manually decode one example: check dataset[0]['audio'] yields bytes and plays back; verify its path exists.
Repair or re-ingest the audio data, then rebuild the dataset.
Confirm you are not feeding a metadata-only dataset (e.g. only paths, no actual files).

Example fix

// before: relative audio paths, wrong cwd
ds[0]['audio']['path']  # 'wavs/0001.wav' — missing from cwd
// after: absolute or verified paths
import os
assert os.path.exists(ds[0]['audio']['path'])

Defensive patterns

Strategy: validation

Validate before calling

def snac_examples_valid(dataset, n=3) -> bool:
    for i in range(min(n, len(dataset))):
        a = dataset[i]['audio']
        if not (a.get('array') is not None and len(a['array']) > 0 and dataset[i]['text']):
            return False
    return True

assert snac_examples_valid(dataset), "examples fail SNAC input requirements"

Try / catch

try:
    ds = trainer._preprocess_snac_dataset(dataset, mapping)
except ValueError as e:
    if 'No valid examples after SNAC preprocessing' in str(e):
        # read per-example warnings; repair audio/text data and retry
        ...

Prevention

Spot-check several examples for decodable audio and non-empty transcripts before training.
Keep audio files at stable absolute paths, or embed audio in the dataset.
Log skip reasons prominently during preprocessing so systemic failures are debuggable.

When it happens

Trigger: 100% of examples fail SNAC codec encoding: unreadable/zero-length audio, sample-rate decode failures, transcripts that break tokenization, or the audio column containing path strings to files that no longer exist.

Common situations: Audio files deleted or moved after dataset creation; Audio column storing relative paths resolved from the wrong cwd; corrupt audio encodings; transcripts empty for all rows.

Related errors

AI-assisted analysis of unslothai/unsloth@203007d190 (2026-08-15). Data as JSON: /api/errors/aeb3936ba34d52d7. Report an issue: GitHub.