ZhuofengLi/qwen3.5-9b-nemotron-sft-ckpt200
ZhuofengLi/qwen3.5-9b-nemotron-sft-ckpt200 is an intermediate 9 billion parameter Qwen3.5 model, fine-tuned by ZhuofengLi on the NVIDIA Nemotron-Terminal-Corpus dataset. This model is specifically trained to enhance terminal interaction and agentic capabilities, leveraging a 32K context length. It is optimized for tasks involving multi-step terminal execution, including mathematical, coding, and software engineering challenges.
Loading preview...
Model Overview
This model, ZhuofengLi/qwen3.5-9b-nemotron-sft-ckpt200, is an intermediate supervised fine-tuning (SFT) checkpoint of the Qwen3.5-9B architecture. It was fine-tuned by ZhuofengLi using the ms-swift framework with DeepSpeed ZeRO-3, specifically targeting improved terminal interaction and agentic capabilities.
Key Capabilities & Training
- Enhanced Terminal Interaction: The model is trained on the
nvidia/Nemotron-Terminal-Corpusdataset, which comprises 366k multi-step terminal execution trajectories. - Agentic Task Performance: Optimized for tasks requiring sequential actions and problem-solving within a terminal environment.
- Diverse Task Coverage: The training data includes a variety of tasks such as mathematics, code generation, and software engineering challenges.
- Training Details: Fine-tuned with a learning rate of 2e-5, a global batch size of 64, and a maximum sequence length of 262144 tokens, utilizing BF16 precision across 64 H200 GPUs.
When to Use This Model
This model is particularly well-suited for applications requiring:
- Automated Terminal Operations: Interacting with command-line interfaces or scripting environments.
- Agentic Workflows: Developing AI agents that can execute multi-step plans in a terminal.
- Code and Software Engineering Assistance: Tasks involving code execution, debugging, or software development within a terminal context.
- Mathematical Problem Solving: Handling math-related problems that can be solved through terminal commands or scripts.