Whisper-Large-v3-Turbo-GGML (Platinum Series)

This repository contains the Platinum Series universal GGML release of Whisper-Large-v3-Turbo. Optimized for the whisper.cpp ecosystem, this release provides state-of-the-art multilingual transcription and translation with significantly faster inference speeds than the original Large-v3 architecture.

📦 Available Files & Quantization Details

File Name	Quantization	Size	Accuracy	Recommended For
ggml-large-v3-turbo-f16.bin	FP16	~1.6 GB	100%	Master Reference / Transcription
ggml-large-v3-turbo-q8_0.bin	Q8_0	~850 MB	99.9%	Platinum Reference / High-Fidelity
ggml-large-v3-turbo-q6_k.bin	Q6_K	~700 MB	99.7%	High-Quality Audio Analysis
ggml-large-v3-turbo-q5_k.bin	Q5_K	~600 MB	99.5%	Balanced Desktop Performance
ggml-large-v3-turbo-q4_k.bin	Q4_K	~500 MB	98.9%	Mobile / Edge Transcription

---

🐍 Python Inference (whisper-cpp-python)

To run these engines using Python:

from whisper_cpp_python import Whisper

# Initialize the engine
whisper = Whisper(
    model_path="ggml-large-v3-turbo-q8_0.bin",
    n_gpu_layers=-1 # Target all layers to NVIDIA/Apple GPU
)

# Transcribe audio file
result = whisper.transcribe("meeting_recording.wav")
print(f"Transcription: {result['text']}")

---

💻 For C# / .NET Users (Whisper.net)

This collection is fully compatible with .NET applications via the Whisper.net library, enabling high-performance integration into local automation and assistant projects

using Whisper.net;
using Whisper.net.Ggml;

// Load the high-fidelity Platinum weights
using var factory = WhisperFactory.FromPath("ggml-large-v3-turbo-q8_0.bin");

using var processor = factory.CreateBuilder()
    .WithLanguage("auto")
    .Build();

using var fileStream = File.OpenRead("audio_sample.wav");

await foreach (var result in processor.ProcessAsync(fileStream))
{
    Console.WriteLine($"{result.Start} -> {result.End}: {result.Text}");
}

---

🏗️ Technical Details

• Optimization Tool: whisper.cpp (CUDA-accelerated)
• Architecture: Whisper Large-v3 (Turbo)
• Hardware Validation: Dual-GPU (RTX 3090 + RTX A4000)

---

☕ Support the Forge

Maintaining the production line for localized AI models requires significant hardware resources. If these tools power your research or industrial vision projects, please consider supporting the development:

Platform	Support Link
Global & India	Support via Razorpay

Scan to support via UPI (India Only):

---

Connect with the architect: Abhishek Jaiswal on LinkedIn