You can deploy this model and start earning money today!

paulgCleansheet/
medasr

Name: medasr
Availability: InStock

endpoints_compatible

Views

No views yet

CleansheetLLC/medasr

Custom inference handler for google/medasr on HuggingFace Inference Endpoints.

Why this exists

The default HF Inference Endpoint container bundles Transformers < 5.0.0, which does not support the lasr_ctc architecture used by MedASR. This repo provides a custom handler.py and requirements.txt that installs the correct version.

Deployment

Create a new model on HuggingFace: CleansheetLLC/medasr
Push the contents of this directory to that repo
The model weights are loaded from google/medasr at runtime (no need to copy them)
Create an Inference Endpoint pointing at CleansheetLLC/medasr
- GPU: T4 (minimum)
- Scale-to-zero: enabled
- Task: automatic-speech-recognition
The endpoint will install requirements.txt, then load handler.py

Usage

bash

# Health check
curl -s https://YOUR_ENDPOINT.endpoints.huggingface.cloud \
  -H "Authorization: Bearer $HF_TOKEN"

# Transcribe audio (WAV, 16kHz mono)
curl -X POST https://YOUR_ENDPOINT.endpoints.huggingface.cloud \
  -H "Authorization: Bearer $HF_TOKEN" \
  -H "Content-Type: audio/wav" \
  --data-binary @audio.wav

Audio requirements

Format: WAV (PCM)
Sample rate: 16kHz (handler resamples if different)
Channels: mono (handler converts if stereo)

Response

{"text": "patient presents with chest pain radiating to the left arm"}

paulgCleansheet/medasr