inusann/whisper-medium-q4_0-ggml

Whisper Medium Q4_0 GGML

Unofficial Q4_0 quantized conversion of the OpenAI Whisper Medium model for whisper.cpp-compatible runtimes.

Important notice

This repository is not affiliated with, endorsed by, or maintained by OpenAI or the whisper.cpp project.

This is not an official OpenAI or whisper.cpp model release.

Model details

• Base model: OpenAI Whisper Medium
• Base model identifier: openai/whisper-medium
• Source GGML file: ggml-medium.bin
• Output file: ggml-medium-q4_0.bin
• Quantization format: Q4_0
• Conversion tool: whisper.cpp quantization tool
• Intended runtimes: whisper.cpp and compatible Whisper.net runtimes
• License: MIT

Quantization process

The model was generated from the non-quantized ggml-medium.bin.

It was not generated by re-quantizing ggml-medium-q5_0.bin.

Example command:

quantize.exe ggml-medium.bin ggml-medium-q4_0.bin q4_0

The exact executable name or arguments may differ depending on the whisper.cpp version.

File integrity

• File: ggml-medium-q4_0.bin
• Size: 444493363 bytes
• SHA-256: 6C8FED17FCD5635C2BEBE863E256556D51DB335147F2E968349C69A9D0FB8E65

Usage

Example direct download URL:

https://huggingface.co/inusann/whisper-medium-q4_0-ggml/resolve/main/ggml-medium-q4_0.bin?download=true

Example with curl:

curl.exe -L --fail --retry 5 ^
  -o "ggml-medium-q4_0.bin" ^
  "https://huggingface.co/inusann/whisper-medium-q4_0-ggml/resolve/main/ggml-medium-q4_0.bin?download=true"

After downloading, verify the SHA-256 hash before use.

Compatibility

This model is intended for runtimes that support the GGML/Q4_0 model format.

Compatibility may vary depending on:

• whisper.cpp version
• Whisper.net version
• native runtime version
• CPU instruction support
• model loader implementation

A successful download does not guarantee compatibility with every runtime.

Accuracy and performance

Quantization can reduce memory usage and improve inference speed, but may also change transcription accuracy.

Users should compare this model with the original or other quantizations using the same audio, runtime settings, thread count, and hardware.

Limitations

This repository only provides a quantized model file.

It does not include:

• a hosted inference API
• guaranteed runtime compatibility
• accuracy guarantees
• performance guarantees
• technical support from OpenAI or whisper.cpp

License and attribution

The underlying OpenAI Whisper project is distributed under the MIT License.

The whisper.cpp project used for quantization is also distributed under the MIT License.

See:

• LICENSE
• LICENSE-OPENAI-WHISPER
• LICENSE-WHISPER-CPP
• THIRD_PARTY_NOTICES.md

Disclaimer

THE MODEL IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND.

Users are responsible for verifying suitability, accuracy, security, legal requirements, and runtime compatibility for their own use cases.