llm type_error ai_generated true

ValueError: Pooling mode 'mean' not supported for this model. Expected 'cls' pooling.

ID: llm/embedding-model-mismatch-pooling

Also available as: JSON · Markdown · 中文
80%Fix Rate
85%Confidence
1Evidence
2024-02-15First Seen

Version Compatibility

VersionStatusIntroducedDeprecatedNotes
sentence-transformers 2.2.0 active — — —
sentence-transformers 2.3.0 active — — —
transformers 4.35.0 active — — —

Root Cause

Using a sentence-transformer model with an incompatible pooling method during embedding generation, often due to model card misreading.

generic

中文

在生成嵌入时使用了与句子变换器模型不兼容的池化方法，通常因误读模型卡导致。

Official Documentation

https://www.sbert.net/docs/pretrained_models.html

Workarounds

  1. 95% success Change pooling mode to 'cls' explicitly: model = SentenceTransformer('model-name', device='cpu'); embeddings = model.encode(sentences, pool='cls') 
    Change pooling mode to 'cls' explicitly: model = SentenceTransformer('model-name', device='cpu'); embeddings = model.encode(sentences, pool='cls')
  2. 85% success Use a different model that supports 'mean' pooling, e.g., 'all-MiniLM-L6-v2' 
    Use a different model that supports 'mean' pooling, e.g., 'all-MiniLM-L6-v2'

中文步骤

  1. Change pooling mode to 'cls' explicitly: model = SentenceTransformer('model-name', device='cpu'); embeddings = model.encode(sentences, pool='cls')
  2. Use a different model that supports 'mean' pooling, e.g., 'all-MiniLM-L6-v2'

Dead Ends

Common approaches that don't work:

  1. Upgrading sentence-transformers to latest version without checking model compatibility 90% fail

    Pooling mode is model-specific, not version-specific; upgrade doesn't change model internals.

  2. Setting pooling mode to 'auto' hoping library will infer correctly 70% fail

    Auto-detection may default to unsupported mode if model config is ambiguous.