OpenBMB/ChatDev · error · ValueError

LocalEmbedding requires model_path parameter

Error message

LocalEmbedding requires model_path parameter

What it means

LocalEmbedding requires params.model_path to point at a local sentence-transformers model. Without it there is nothing to load, so __init__ raises before loading the model.

Source

Thrown at runtime/node/agent/memory/embedding.py:171

        elif self.chunk_strategy == 'weighted':
            # Weighted aggregation (earlier chunks weigh more)
            weights = [1.0 / (i + 1) for i in range(len(chunk_embeddings))]
            total_weight = sum(weights)
            return [sum(chunk[i] * weights[j] for j, chunk in enumerate(chunk_embeddings)) / total_weight 
                   for i in range(len(chunk_embeddings[0]))]
        else:
            # Default to the first chunk
            return chunk_embeddings[0]

class LocalEmbedding(EmbeddingBase):
    def __init__(self, embedding_config: EmbeddingConfig):
        super().__init__(embedding_config)
        self.model_path = embedding_config.params.get('model_path')
        self.device = embedding_config.params.get('device', 'cpu')
        self._fallback_dim = 768  # Default; updated after first successful call
        
        if not self.model_path:
            raise ValueError("LocalEmbedding requires model_path parameter")
        
        # Load the local embedding model (e.g., sentence-transformers)
        try:
            from sentence_transformers import SentenceTransformer
            self.model = SentenceTransformer(self.model_path, device=self.device)
        except ImportError:
            raise ImportError("sentence-transformers is required for LocalEmbedding")

    def get_embedding(self, text):
        # Preprocess text before encoding
        processed_text = self._preprocess_text(text)
        
        if not processed_text:
            return [0.0] * self._fallback_dim
        
        try:
            embedding = self.model.encode(processed_text, convert_to_tensor=False)
            result = embedding.tolist()

View on GitHub (pinned to 4fb2db0ea9)

Solutions

1. Add params={'model_path': '/path/to/model-or-repo-id'} to the embedding config
2. Use a valid sentence-transformers model id (e.g. 'sentence-transformers/all-MiniLM-L6-v2')
3. Verify the key is exactly 'model_path' (snake_case)

Example fix

# before
EmbeddingConfig(provider='local', params={'device': 'cpu'})

# after
EmbeddingConfig(provider='local', params={'model_path': 'sentence-transformers/all-MiniLM-L6-v2', 'device': 'cpu'})

if embedding_config.provider == 'local' and not embedding_config.params.get('model_path'):
    raise ConfigError('local embedding requires params.model_path')

When it happens

Trigger: provider='local' with no params dict or no 'model_path' key: EmbeddingConfig(provider='local', params={'device':'cpu'}) — device alone is not enough.

Common situations: Assuming 'local' downloads a default model; forgetting to include the path to the downloaded model directory; typo like 'modelPath' or 'path' in params.

Related errors

• Unsupported embedding model: {model}
• sentence-transformers is required for LocalEmbedding

AI-assisted analysis of OpenBMB/ChatDev@4fb2db0ea9 (2026-08-27). Data as JSON: /api/errors/3de41691dc7602f3. Report an issue: GitHub.