You can deploy this model and start earning money today!

xpoon/
medical-whisper-large-v3-ggml

Name: medical-whisper-large-v3-ggml
Availability: InStock

Views

No views yet

Medical Whisper Models (GGML Format)

Collection of medical and clinical Whisper models converted to GGML format for use with whisper.cpp.

Models Available

Na0s Medical-Whisper-Large-v3

Converted from Na0s/Medical-Whisper-Large-v3 - Based on Whisper Large-v3.

File	Size	Precision	Description
`ggml-medical-whisper-large-v3.bin`	2.9 GB	f16	Full precision, best medical accuracy
`ggml-medical-whisper-large-v3-q5_0.bin`	1.0 GB	Q5_0	Quantized, 65% smaller

Best for: General medical terminology, anatomical terms, medications, clinical speech

Knowtex Medical-GovCloud v2

Converted from Knowtex-ai/whisper-medical-govcloud - Based on Whisper Large-v2.

File	Size	Precision	Description
`ggml-knowtex-medical-v2.bin`	2.9 GB	f16	Full precision, best clinical accuracy
`ggml-knowtex-medical-v2-q5_0.bin`	1.0 GB	Q5_0	Quantized, 65% smaller

Best for: Clinical transcription, healthcare/government applications, medical documentation

Model Comparison

Feature	Na0s Medical v3	Knowtex Medical v2
Base Model	Whisper Large-v3	Whisper Large-v2
Training Focus	General medical	Clinical/GovCloud
Architecture	Latest (v3)	Previous (v2)
Use Case	Medical terminology	Clinical transcription

Usage with whisper.cpp

Command Line

bash

# Download whisper.cpp
git clone https://github.com/ggml-org/whisper.cpp
cd whisper.cpp && make

# Download a model (replace MODEL_NAME)
wget https://huggingface.co/xpoon/medical-whisper-large-v3-ggml/resolve/main/MODEL_NAME

# Run inference
./build/bin/whisper-cli -m MODEL_NAME -f audio.wav -l en

Recommended Model Selection

For iOS/Mobile:

Use Q5_0 quantized versions (1.0 GB each)
~1.5 GB runtime memory
Minimal quality loss (<1% WER)

For Desktop/Server:

Use full precision versions (2.9 GB each)
~3.9 GB runtime memory
Best possible accuracy

For Medical Terminology:

Try Na0s models first
Newer architecture (v3) may perform better on general medical speech

For Clinical Documentation:

Try Knowtex models
Specialized for clinical/healthcare transcription

Performance

Speed: ~6x faster than Python Whisper
Platforms: macOS, iOS, Linux, Windows, Android
GPU Acceleration: Metal (Apple Silicon), CUDA, OpenCL
Memory Efficient: Quantized models use 65% less RAM

Conversion Details

All models converted using whisper.cpp tools:

Downloaded from HuggingFace (SafeTensors format)
Converted using convert-h5-to-ggml.py
Quantized using whisper.cpp quantize tool

Original Models

Na0s Medical-Whisper-Large-v3: https://huggingface.co/Na0s/Medical-Whisper-Large-v3
Knowtex Medical-GovCloud: https://huggingface.co/Knowtex-ai/whisper-medical-govcloud

Credits

Original models: Na0s, Knowtex-ai
Base model: OpenAI Whisper
Conversion tools: whisper.cpp
GGML framework: ggml

License

Same licenses as the original models. Refer to original repositories for licensing details.

Last Updated: December 2024