Update README.md

README.md

@@ -1,74 +1,208 @@
-# Qwen3-4B-Computer-Science Models
+---
+language:
+  - en
+license: apache-2.0
+base_model: Irfanuruchi/Qwen3-4B-Computer-Science
+base_model_relation: quantized
+library_name: optimum-intel
+pipeline_tag: text-generation
+inference: false
+tags:
+  - qwen3
+  - qwen3-for-causal-lm
+  - computer-science
+  - software-engineering
+  - programming
+  - coding
+  - openvino
+  - optimum-intel
+  - int4
+  - quantized
+  - cpu
+  - transformers
+---
+
+# Qwen3-4B-Computer-Science OpenVINO INT4
+
+This repository provides an **OpenVINO INT4** version of **Qwen3-4B-Computer-Science**, optimized for efficient inference using **Intel OpenVINO** and **Optimum Intel**.
+
+The model is optimized for software engineering and computer science workloads while significantly reducing storage requirements and memory usage through INT4 weight compression, enabling efficient CPU inference on OpenVINO-supported hardware.
 
-Official model releases for the **Qwen3-4B-Computer-Science** project.
+---
 
-This collection provides officially maintained releases of the model across multiple inference runtimes and hardware platforms, allowing users to choose the format best suited to their deployment environment.
+# Model Information
+
+| Property | Value |
+|----------|-------|
+| Base Model | Irfanuruchi/Qwen3-4B-Computer-Science |
+| Base Architecture | Qwen3 |
+| Parameters | ~4 Billion |
+| Model Format | OpenVINO IR |
+| Weight Compression | INT4 Asymmetric |
+| Compression Group Size | 128 |
+| Runtime | OpenVINO Runtime |
+| Library | Optimum Intel |
+| Primary Hardware | CPU |
+| License | Apache License 2.0 |
 
 ---
 
-## Available Formats
+# Features
 
-| Format | Status | Primary Runtime |
-|---------|:------:|-----------------|
-| BF16 | ✅ | Transformers |
-| GGUF | ✅ | llama.cpp, Ollama, LM Studio |
-| AWQ | ✅ | Compressed-Tensors compatible runtimes |
-| MLX 4-bit | ✅ | Apple MLX |
-| MLX 8-bit | ✅ | Apple MLX |
-| MLX BF16 | ✅ | Apple MLX |
-| OpenVINO INT4 | ✅ | OpenVINO + Optimum Intel |
+- Optimized OpenVINO IR format
+- INT4 asymmetric weight compression
+- Reduced storage footprint
+- Lower memory usage
+- CPU-optimized inference
+- Compatible with Hugging Face Transformers through Optimum Intel
+- Exported tokenizer and detokenizer
+- Chat template included
+- Ready for local deployment
 
 ---
 
-## Included Models
+# Export Configuration
 
-### Qwen3-4B-Computer-Science
-Original BF16 release intended for Hugging Face Transformers, fine-tuning, and general inference.
+The model was exported using:
 
-### Qwen3-4B-Computer-Science-GGUF
-GGUF release for efficient inference with llama.cpp-compatible runtimes such as Ollama and LM Studio.
+```bash
+optimum-cli export openvino \
+    --model Irfanuruchi/Qwen3-4B-Computer-Science \
+    --task text-generation-with-past \
+    --weight-format int4 \
+    Qwen3-4B-Computer-Science-OpenVINO-INT4
+```
 
-### Qwen3-4B-Computer-Science-AWQ
-Activation-Aware Weight Quantization (AWQ) release using the Compressed-Tensors format for supported inference frameworks.
+Compression summary:
 
-### Qwen3-4B-Computer-Science-MLX-4bit
-Optimized 4-bit MLX release for Apple Silicon devices.
+- **252 transformer layers** compressed using **INT4 asymmetric** with **group size 128**
+- **1 auxiliary layer** stored using **INT8 per-channel**
 
-### Qwen3-4B-Computer-Science-MLX-8bit
-Optimized 8-bit MLX release providing a balance between quality and memory usage.
+---
 
-### Qwen3-4B-Computer-Science-MLX-BF16
-Full-precision BF16 MLX release for maximum inference quality on Apple Silicon.
+# Installation
 
-### Qwen3-4B-Computer-Science-OpenVINO-INT4
-OpenVINO INT4 release optimized for efficient CPU inference using Intel OpenVINO and Optimum Intel.
+```bash
+pip install -U openvino optimum-intel transformers
+```
 
 ---
 
-## Purpose
+# Example Usage
+
+```python
+from transformers import AutoTokenizer
+from optimum.intel.openvino import OVModelForCausalLM
+
+model_id = "Irfanuruchi/Qwen3-4B-Computer-Science-OpenVINO-INT4"
+
+tokenizer = AutoTokenizer.from_pretrained(model_id)
 
-The objective of this collection is to provide officially maintained releases of the same model across multiple deployment formats while preserving consistent behavior and compatibility with their respective inference ecosystems.
+model = OVModelForCausalLM.from_pretrained(
+    model_id,
+    device="CPU",
+)
 
-Each repository includes:
+messages = [
+    {
+        "role": "system",
+        "content": "You are a computer science assistant."
+    },
+    {
+        "role": "user",
+        "content": "Explain Floyd's cycle detection algorithm."
+    },
+]
 
-- Documentation
-- Runtime-specific usage examples
-- License information
-- SHA256 integrity verification files
-- Configuration files required for the target runtime
+prompt = tokenizer.apply_chat_template(
+    messages,
+    tokenize=False,
+    add_generation_prompt=True,
+)
+
+inputs = tokenizer(prompt, return_tensors="pt")
+
+outputs = model.generate(
+    **inputs,
+    max_new_tokens=256,
+    do_sample=False,
+)
+
+response = tokenizer.decode(
+    outputs[0][inputs["input_ids"].shape[1]:],
+    skip_special_tokens=True,
+)
+
+print(response)
+```
 
 ---
 
-## Model Family
+# Validation
+
+The exported model was successfully validated using:
+
+- OpenVINO 2026.2.1
+- CPU inference
+- Chat template support
+- Greedy decoding
+- Software engineering benchmark prompts
+
+Validation confirmed successful generation of technically correct programming responses, including algorithm implementation and complexity analysis.
+
+---
+
+# Intended Use
+
+This model is intended for:
+
+- Software engineering assistance
+- Computer science education
+- Programming support
+- Code generation
+- Code review
+- Debugging
+- Algorithm design
+- Technical documentation
+- Technical question answering
+
+---
+
+# Model Family
+
+The **Qwen3-4B-Computer-Science** release family currently includes:
+
+| Model | Status |
+|------|:------:|
+| Transformers BF16 | ✅ |
+| GGUF (Multiple Quantization Variants) | ✅ |
+| AWQ | ✅ |
+| MLX 4-bit | ✅ |
+| MLX 8-bit | ✅ |
+| MLX BF16 | ✅ |
+| OpenVINO INT4 | ✅ |
+
+Each release is maintained in its own repository with runtime-specific documentation, usage examples, integrity verification files, and configuration for its respective inference backend.
+
+---
+
+# Limitations
+
+Like other large language models, this model may occasionally:
+
+- Generate incorrect or incomplete code
+- Hallucinate APIs or implementation details
+- Produce inefficient implementations
+- Misinterpret ambiguous instructions
+
+All generated code should be reviewed and tested before use in production or safety-critical environments.
+
+INT4 weight compression may introduce minor quality differences compared with higher-precision model variants.
+
+---
 
-This release family currently includes:
+# License
 
-- ✅ Transformers BF16
-- ✅ GGUF
-- ✅ AWQ
-- ✅ MLX 4-bit
-- ✅ MLX 8-bit
-- ✅ MLX BF16
-- ✅ OpenVINO INT4
+This repository is distributed under the **Apache License 2.0**.
 
-Additional deployment formats may be added in future releases as the ecosystem evolves.
+See the included **LICENSE** file for the complete license text and applicable attribution requirements.