anmoldhandhania93/ANMOLGPT-4B-v0.3
ANMOLGPT-4B-v0.3 is a 4.5 billion parameter causal language model developed by Anmol Dhandhania, built upon Qwen3.5-4B. This version is fine-tuned on the TruthfulQA dataset using LoRA with Unsloth, specifically to enhance factual accuracy and reasoning capabilities. It demonstrates measurable improvements in reasoning benchmarks like ARC-Easy, TruthfulQA MC2, and MMLU compared to its predecessor, while maintaining strong general-purpose performance. The model is intended for applications requiring improved truthfulness and reasoning, such as conversational AI, question answering, and research assistance.
Loading preview...
ANMOLGPT-4B-v0.3: Enhanced Factual Accuracy and Reasoning
ANMOLGPT-4B-v0.3 is the third iteration in the ANMOLGPT series, developed by Anmol Dhandhania. This 4.5 billion parameter model is built on the robust Qwen3.5-4B architecture and has undergone targeted fine-tuning using LoRA with Unsloth on the TruthfulQA dataset. The primary focus of this version is to significantly improve factual accuracy and reasoning performance.
Key Capabilities & Improvements
- Enhanced Factual Accuracy: Direct fine-tuning on TruthfulQA leads to more reliable and truthful responses.
- Improved Reasoning Performance: Demonstrates measurable gains across various reasoning benchmarks, including ARC-Easy (+2.27), TruthfulQA MC2 (+3.33), and MMLU (+7.74) compared to v0.2.
- Strong General-Purpose Abilities: Maintains solid performance in general language understanding tasks, with benchmark scores like HellaSwag (72.80 Norm) and PIQA (78.45 Norm).
- Open-Source & Apache 2.0: Provides flexibility for developers and researchers.
Ideal Use Cases
ANMOLGPT-4B-v0.3 is well-suited for applications where factual correctness and logical reasoning are critical:
- Conversational AI: For more accurate and reliable dialogue systems.
- Question Answering: Providing precise and truthful answers.
- Reasoning Tasks: Excelling in complex problem-solving scenarios.
- Educational Applications: Supporting learning with factual information.
- Research Assistance: Aiding in information retrieval and synthesis.
- AI Experimentation: A robust base for further research and development.