Gemma-2-2B-IT-OpenVINO-INT8
This repository contains the Silver Series optimized OpenVINOβ’ IR version of Gemma-2-2B-Instruct, quantized to INT8 precision using NNCF. This release offers a high-fidelity balance between reasoning accuracy and local execution speed.
π Python Inference (Optimum-Intel)
To run this Silver Series engine locally:
from optimum.intel import OVModelForCausalLM
from transformers import AutoTokenizer
model_id = "CelesteImperia/Gemma-2-2B-IT-OpenVINO-INT8"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = OVModelForCausalLM.from_pretrained(model_id)
prompt = "Explain the benefit of INT8 quantization for AI models."
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
π» For C# / .NET Users (OpenVINO.GenAI)
The Silver Series is optimized for the native OpenVINO.GenAI NuGet package. This approach provides the lowest latency for Windows-based C# applications.
using OpenVino.GenAI;
// 1. Initialize the LLM Pipeline
var device = "CPU"; // Use "GPU" for RTX 3090/A4000 acceleration
using var pipe = new LLMPipeline("path/to/gemma-2-int8-model", device);
// 2. Set Generation Config
var config = new GenerationConfig { MaxNewTokens = 256, Temperature = 0.7f };
// 3. Execute Inference
var prompt = "Translate this C# snippet to Python: Console.WriteLine(\"Forge Active\");";
var result = pipe.Generate(prompt, config);
Console.WriteLine(result);
ποΈ Technical Details
- Optimization Tool: NNCF (Neural Network Compression Framework)
- Quantization: INT8 Symmetric (Per-channel)
- Series: Silver (High Fidelity)
- Workstation Validation: Dual-GPU (RTX 3090 + RTX A4000)
β Support the Forge
Maintaining a comprehensive local AI warehouse requires significant resources. If these high-fidelity Silver Series models power your projects, please consider supporting the Forge:
| Platform | Support Link |
|---|---|
| Global & India | Support via Razorpay |
Scan to support via UPI (India Only):
π License
This model is released under the Gemma Terms of Use.
Connect with the architect: Abhishek Jaiswal on LinkedIn
- Downloads last month
- 16