whisper
I built docker using official github runpod and i get 024-08-08T20:20:40.508212050Z {"requestId": null, "message": "{\n "error_type": "<class 'RuntimeError'>",\n "error_message": "Library libcublas.so.12 is not found or cannot be loaded",\n "error_traceback": "Traceback (most recent call last):\n File \"/usr/local/lib/python3.10/dist-packages/runpod/serverless/modules/rp_job.py\", line 134, in run_job\n handler_return = handler(job)\n File \"/usr/local/lib/python3.10/dist-packages/runpod/serverless/utils/rp_debugger.py\", line 165, in call\n result = self.function(*args, **kwargs)\n File \"/rp_handler.py\", line 72, in run_whisper_job\n whisper_results = MODEL.predict(\n File \"/predict.py\", line 76, in predict\n segments, info = list(model.transcribe(str(audio),\n File \"/usr/local/lib/python3.10/dist-packages/faster_whisper/transcribe.py\", line 344, in transcribe\n encoder_output = self.encode(segment)\n File \"/usr/local/lib/python3.10/dist-packages/faster_whisper/transcribe.py\", line 767, in encode\n return self.model.encode(features, to_cpu=to_cpu)\nRuntimeError: Library libcublas.so.12 is not found or cannot be loaded\n",\n "hostname": "08h82mdvx3cl2d-64411542",\n "worker_id": "08h82mdvx3cl2d",\n "runpod_version": "1.6.2"\n}", "level": "ERROR"} here is the git i used https://github.com/runpod-workers/worker-faster_whisper
GitHub
GitHub - runpod-workers/worker-faster_whisper: 🎧 | RunPod worker of...
🎧 | RunPod worker of the faster-whisper model for Serverless Endpoint. - runpod-workers/worker-faster_whisper
1 Reply
I suggest you modify the Dockerfile from this:
FROM nvidia/cuda:11.7.1-cudnn8-runtime-ubuntu20.04
to this:
FROM nvidia/cuda:11.8-base-ubuntu20.04
This should resolve your missing libcublas.so.12 library. You will need to rebuild the image.
This issue can surface when you build an image that is configured to import modules that uses an unspecified version. This directs the code to use the latest version of modules. When this happens your build is no longer the image that your code expects. This can cause incompatibilities. With the way that modules are nested within modules in Python it is very difficult to maintain specific versioning if your code uses any modules. The only way you can keep it locked into a specific configuration is in image form.