unslothai/unsloth · error · ValueError

Unknown format_type: {format_type}

Error message

Unknown format_type: {format_type}

What it means

ValueError from the dataset standardization dispatcher in dataset_utils: format_type did not match any handled branch. The function dispatches on format_type — 'raw', 'auto', 'alpaca', and 'chatml'/'conversational'/'sharegpt' — and the final else raises rather than guessing, since an unknown target format would silently produce the wrong schema. Auto-detected but unrecognized DATA returns a graceful 'unknown' result dict; only an unknown format_type ARGUMENT raises.

Source

Thrown at studio/backend/utils/datasets/dataset_utils.py:862

                        "warnings": warnings,
                    }
                except Exception as e:
                    warnings.append(f"Standardization failed: {e}")

            return {
                "dataset": dataset,
                "detected_format": "unknown",
                "final_format": "unknown",
                "chat_column": detected["chat_column"],
                "is_standardized": False,
                "requires_manual_mapping": True,
                "is_image": multimodal_info["is_image"],
                "multimodal_info": multimodal_info,
                "warnings": warnings,
            }

    else:
        raise ValueError(f"Unknown format_type: {format_type}")


def format_and_template_dataset(
    dataset,
    model_name,
    tokenizer,
    is_vlm = False,
    format_type = "auto",
    # VLM-specific parameters
    vlm_instruction = None,  # Now optional - will auto-generate
    vlm_text_column = None,
    vlm_image_column = None,
    dataset_name = None,
    custom_prompt_template = None,
    add_eos_token = False,
    remove_bos_prefix = False,
    custom_format_mapping = None,
    auto_detect_custom = True,

View on GitHub (pinned to 203007d190)

Solutions

1. Use one of the supported values: 'auto', 'raw', 'alpaca', 'chatml', 'conversational', or 'sharegpt' (lowercase)
2. Default to format_type='auto' when you want the pipeline to detect the layout itself
3. Validate format_type against the supported set at the config/UI boundary before invoking the pipeline

Example fix

# before
result = format_and_template_dataset(ds, model, tok, format_type='ChatML')  # raises

# after
SUPPORTED = {'auto', 'raw', 'alpaca', 'chatml', 'conversational', 'sharegpt'}
assert format_type in SUPPORTED, f"format_type must be one of {SUPPORTED}"
result = format_and_template_dataset(ds, model, tok, format_type=format_type.lower())

SUPPORTED_FORMAT_TYPES = {"auto", "raw", "alpaca", "chatml", "conversational", "sharegpt"}

def is_supported_format_type(format_type: str) -> bool:
    return format_type in SUPPORTED_FORMAT_TYPES

def normalize_format_type(value: str) -> str:
    v = (value or "auto").strip().lower()
    if v not in {"auto", "raw", "alpaca", "chatml", "conversational", "sharegpt"}:
        raise ValueError(f"unsupported format_type: {value!r}")
    return v

When it happens

Trigger: Calling format_and_template_dataset(..., format_type='ChatML') (wrong case), 'instruct', 'json', or any string outside the supported set listed above.

Common situations: Config/UI free-text format fields forwarded unvalidated; casing mismatches ('Alpaca' vs 'alpaca'); renamed format values after upgrading the library; mixing up dataset format names with training argument names.

Related errors

• No 'messages' or 'conversations' or 'texts' column found.
• Add a Provider connection block before running this recipe.
• Unsupported attention_backend '{value}'. Use one of: {', '.j
• Unsupported transformer_cache '{value}'. Use one of: off, au
• too many extra llama-server args (limit {MAX_EXTRA_ARG_TOKEN

AI-assisted analysis of unslothai/unsloth@203007d190 (2026-08-15). Data as JSON: /api/errors/524227a76e630036. Report an issue: GitHub.