You can use the provided script to download the medium model: bash ./models/download-ggml-model.sh medium Use code with caution.
Here's a step-by-step guide to getting up and running on your own machine.
echo "Running inference..." ./main -m $MODEL_FILE -p "What is the capital of France?" -n 50
This public link is valid for 7 days and shares a thread, including any personal information you added. This link or copies made by others cannot be deleted. If you share with third parties, their policies apply. Can’t copy the link right now. Try again later.
Here are the most common quantization types you will encounter, along with their key characteristics: ggmlmediumbin work
refers to the compiled weight file for the "Medium" variant of OpenAI’s Whisper automatic speech recognition (ASR) model, specifically formatted for use with the whisper.cpp library. Technical Overview
.bin is a raw binary file containing the model weights. Unlike .safetensors (which has metadata headers), .bin files are often memory-mapped directly, allowing near-instantaneous loading.
Use the provided script: sh ./models/download-ggml-model.sh medium . Compile: Build the project using cmake or make . Run: Execute the transcription via command line: ./main -m models/ggml-medium.bin -f your_audio.wav Use code with caution. Copied to clipboard If you'd like, I can help you:
./build/bin/whisper-cli -m models/ggml-model-q5_0.bin -f audio.wav You can use the provided script to download
: Developed by Georgi Gerganov , GGML is the engine that allows these models to run efficiently on standard hardware without heavy GPU requirements. You can explore the technical implementation details in the Introduction to GGML on Hugging Face.
GGML is a tensor library for machine learning designed for large models and . Unlike PyTorch or TensorFlow (which are GPU-centric), GGML is optimized for Apple Silicon (M1/M2/M3), ARM64, and x86 CPUs with AVX2 support. It enables running quantized LLMs on consumer hardware without a dedicated GPU.
./quantize original-f32.bin model.q5_1.bin q5_1
Working with a (e.g., 13B parameters) stored as a .bin file. This link or copies made by others cannot be deleted
Unlike Tiny or Base , the Medium model has the deep context understanding required to translate accurately across dozens of different languages and dialects. 3. How whisper.cpp Processes ggml-medium.bin
A single .bin file contains all the tensor weights, model configuration layouts, and vocabulary rules needed for speech tokenization.
Thanks to the GGML architecture, the workload isn’t restricted solely to your computer's processor. You can offload parts of the workload to Apple Silicon (Metal), NVIDIA/AMD GPUs (using CUDA/OpenCL), or even integrate OpenVINO for certain processors. 5. Getting Started: How It Works in Practice