Phi-3.5-mini-instruct-OpenVINO-INT8 (Silver Series)

This repository contains the Silver Series optimized OpenVINO™ IR version of Phi-3.5-mini-instruct, quantized to INT8 precision using NNCF. This high-fidelity release maximizes the model's exceptional reasoning and mathematical capabilities while ensuring rapid local inference.

---

🐍 Python Inference (Optimum-Intel)

To run this Silver Series engine locally using the optimum-intel library:

from optimum.intel import OVModelForCausalLM
from transformers import AutoTokenizer

model_id = "CelesteImperia/Phi-3.5-mini-instruct-OpenVINO-INT8"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = OVModelForCausalLM.from_pretrained(model_id)

prompt = "Solve this step-by-step: If 3x + 5 = 20, what is x?"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=150)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

---

💻 For C# / .NET Users (OpenVINO.GenAI)

The Silver Series is optimized for native OpenVINO.GenAI NuGet integration, offering an efficient path for embedding high-logic agents into Windows software.

using OpenVino.GenAI;

// 1. Initialize the LLM Pipeline
var device = "CPU"; // Use "GPU" for RTX acceleration
using var pipe = new LLMPipeline("path/to/phi-3.5-int8-model", device);

// 2. Set Generation Config
var config = new GenerationConfig { MaxNewTokens = 300, Temperature = 0.5f };

// 3. Execute Inference
var prompt = "Explain the difference between an Interface and an Abstract Class in C#.";
var result = pipe.Generate(prompt, config);

Console.WriteLine(result);

---

🏗️ Technical Details

• Optimization Tool: NNCF (Neural Network Compression Framework)
• Quantization: INT8 Symmetric (Per-channel)
• Primary Task: Logic / Technical Reasoning
• Workstation Validation: Dual-GPU (RTX 3090 + RTX A4000)

---

☕ Support the Forge

Maintaining the production line for high-fidelity models requires significant power and hardware resources. If these tools power your development, please consider supporting the Forge:

Platform	Support Link
Global & India	Support via Razorpay

Scan to support via UPI (India Only):

---

📜 License

This model is released under the MIT License.

---

Connect with the architect: Abhishek Jaiswal on LinkedIn