oscerd opened a new pull request, #25899: URL: https://github.com/apache/camel/pull/25899
## Issue [CAMEL-24527](https://issues.apache.org/jira/browse/CAMEL-24527) ## Problem The `authToken` option (and the token resolved from `oauthProfile`) was honoured by **only** the chat task. `ChatPredictor.getPythonScript()` builds a `token='...'` clause and passes it to `transformers.pipeline()`; none of the other nine task predictors do. As a result, text generation, summarization, question answering, classification, sentence embeddings, ASR, TTS, text-to-image and zero-shot classification all failed with HTTP 401 when loading a gated or private model, even with a token configured. ## Fix Apply the token centrally in `AbstractTaskPredictor.loadModel()` by exporting it as the standard `HF_TOKEN` environment variable at the top of the generated handler script: ```python import os os.environ['HF_TOKEN'] = '<token>' ``` `transformers` / `huggingface_hub` read `HF_TOKEN` from the environment when no explicit token is passed, so this authenticates **every** task in a single place instead of threading a token clause through nine predictors and nine Python templates. The token is read from `config.getAuthToken()`, which holds either the `authToken` option or the value resolved from an OAuth profile (`HuggingFaceProducer.resolveOAuthToken`). When no token is configured, the script is unchanged. ## Testing - New `AuthTokenInjectionTest` verifies the token is exported as `HF_TOKEN` when configured and that the script is left untouched when it is not. - `mvn -Psourcecheck validate` green. _Claude Code on behalf of oscerd_ -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
