calesthio/OpenMontage · error · ValueError

Unknown model: {model_name}

Error message

Unknown model: {model_name}

What it means

Raised at the end of _load_model when model_name matches none of the supported vision models (the code shows branches for 'clip' and 'llava' preceding the raise). It is a pure input-validation error: the string was not validated against the allowlist before model loading was attempted. Fix is to pass one of the supported model names.

Source

Thrown at tools/analysis/video_understand.py:399

            return model, processor, device

        if model_name == "blip2":
            model_id = "Salesforce/blip2-opt-2.7b"
            processor = Blip2Processor.from_pretrained(model_id)
            model = Blip2ForConditionalGeneration.from_pretrained(
                model_id, torch_dtype=torch.float16 if device == "cuda" else torch.float32
            ).to(device)
            return model, processor, device

        if model_name == "llava":
            model_id = "llava-hf/llava-1.5-7b-hf"
            processor = AutoProcessor.from_pretrained(model_id)
            model = AutoModelForCausalLM.from_pretrained(
                model_id, torch_dtype=torch.float16 if device == "cuda" else torch.float32
            ).to(device)
            return model, processor, device

        raise ValueError(f"Unknown model: {model_name}")

    def _analyze_describe(
        self, frames: list, model_name: str
    ) -> list[dict[str, Any]]:
        """Generate captions for each frame."""
        import torch

        model, processor, device = self._load_model(model_name)
        results = []

        for i, img in enumerate(frames):
            if model_name == "clip":
                # CLIP is not a captioning model; use zero-shot classification
                # with generic scene descriptions as a caption proxy
                candidate_texts = [
                    "a photo of a person", "a photo of a landscape",
                    "a photo of an object", "a photo of text",
                    "a photo of an animal", "a photo of a building",

View on GitHub (pinned to 95e1c3d0ab)

Solutions

1. Pass exactly one of the supported names (check the branches in _load_model — e.g. 'clip' or 'llava')
2. Validate model_name against the allowlist before calling the tool and fail fast with a helpful message listing valid options
3. Normalize input with model_name.strip().lower() at the boundary if casing is the issue

Example fix

// before
results = tool.run({"model": "GPT-4V"})

// after
ALLOWED = {"clip", "llava"}
model = inputs.get("model", "clip").strip().lower()
if model not in ALLOWED:
    raise ValueError(f"model must be one of {sorted(ALLOWED)}, got {model!r}")
results = tool.run({"model": model})

ALLOWED_MODELS = {"clip", "llava"}  # mirror the branches in _load_model
model = inputs.get("model", "clip")
if model not in ALLOWED_MODELS:
    raise ValueError(f"model must be one of {sorted(ALLOWED_MODELS)}")

def is_supported_model(name: str) -> bool:
    return name in {"clip", "llava"}

When it happens

Trigger: Calling the video-understand describe/analyze tool with model_name values like 'gpt-4v', 'qwen-vl', 'blip', or a typo such as 'CLIP' (case-sensitive) or 'llava1.5'; a config file or CLI flag carrying a model identifier added for a different subsystem.

Common situations: Copied model names from a different tool's docs; case mismatch ('Clip' vs 'clip'); version drift after the tool's supported model list changed; default config value not updated after renaming.

Related errors

• Unknown profile {name!r}. Available: {available}
• ${res.status} ${url}
• fetch failed ${r.status}: ${url}
• path escapes project
• media not found

AI-assisted analysis of calesthio/OpenMontage@95e1c3d0ab (2026-08-15). Data as JSON: /api/errors/7982fa79317e00f6. Report an issue: GitHub.