429 thrown while requesting HEAD https://huggingface.co

[arnm]

@Davidyz Have you ran into this before? Happens when running vectorcode vectorize. First happened on pre commit message so it wasnt clear what the issue was since there was no error log. Had to drop and then vectorize from scratch to see the error message. Its for a project with 25 files as listed by vectorcode ls

HTTP Error 429 thrown while requesting HEAD https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2/resolve/main/modules.json
WARNING: huggingface_hub.utils._http : HTTP Error 429 thrown while requesting HEAD https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2/resolve/main/modules.json

[Davidyz]

Hi, did this happen the first time you used it, or one of the subsequent runs? 429 error seems to indicate that there were too many requests sent from your device to the huggingface server. You can try add the following in your VectorCode config:

{
  "embedding_params": {
    "model_name": "sentence-transformers/all-MiniLM-L6-v2",
    "local_files_only": true
  },
  "reranker": "CrossEncoderReranker",
  "reranker_params": {
    "local_files_only": true
  },
}

The local_files_only parameter will prevent the sentence transformers from sending excessive (and unnecessary) requests to Huggingface when you've already got the models downloaded on your system. See here for more information.

[arnm]

Thank you, I'll try this in the future. It was on subsequent runs not on the initial setup.

Only thing I would consider would be maybe adding a timeout when requesting from external services, if there isn't already logic for that. It seemed like it didn't time out or log an error when doing the update. Only when I did vectorize did I see the error

[Davidyz]

Only thing I would consider would be maybe adding a timeout when requesting from external services

By default, there's no intention to send any stuff to an external service. I even deliberately turned off the chromadb telemetry. The default embedding library (sentence transformer) sends a one-off request to huggingface to download the model, and that should be it. It seems like sentence transformer still sends some requests to huggingface in subsequent runs to validate the model, but the local_files_only option should be able to prevent those requests. If it doesn't already, I remember seeing a PR in their repo about this, so hopefully this'll be fixed soon.