pankajpandey-dev/Qwen3-4B-Hindi-Instruct-v2
The pankajpandey-dev/Qwen3-4B-Hindi-Instruct-v2 is a 4 billion parameter instruction-tuned language model based on Qwen3-4B, specifically fine-tuned for strong Hindi instruction-following. It is designed to handle both Hindi (Devanagari) and English, making it suitable for bilingual applications. This compact model is optimized for local and edge deployment, offering an openly licensed solution for Hindi chat and assistant applications.
Loading preview...
Overview
pankajpandey-dev/Qwen3-4B-Hindi-Instruct-v2 is a 4 billion parameter language model, part of the Hindi LLM Series, focused on providing robust Indic-language models for local and edge deployment. It is an instruction-tuned version of Qwen3-4B, specifically fine-tuned to follow instructions and respond naturally in Hindi.
Key Capabilities & Features
- Strong Hindi Instruction-Following: Trained on 10,000 curated Hindi instruction-response pairs.
- Bilingual Support: Capable of handling both Hindi (Devanagari) and English.
- Compact Size: With approximately 4 billion parameters, it runs efficiently on a single consumer GPU and quantizes well for CPU deployment.
- Open License: Released under the Apache 2.0 license, allowing for commercial use.
- Efficient Fine-tuning: Fine-tuned using LoRA (r=32, α=32) via Unsloth, with the dataset filtered for genuine Hindi responses.
Intended Use Cases
This model is primarily intended for:
- Hindi chat and assistant applications.
- Instruction-following tasks in Hindi.
- Experimentation with Indic-language models.
- Local and edge deployment, including a GGUF version for CPU-based inference with tools like llama.cpp, Ollama, or LM Studio.
Limitations
As a 4B parameter model, it may exhibit factual errors and inconsistent results on complex reasoning or highly specialized domains. It inherits biases from its base model and training data, so validation of outputs is recommended for production use.