allenai/qwen35-9b-endless
The allenai/qwen35-9b-endless model is a 9 billion parameter language model developed by Ai2, fine-tuned from Qwen 3.5 9B. It is specifically optimized using DPPO for use as a terminal-agent, trained on the endless-terminals dataset. This model excels in terminal-based tasks, demonstrating improved performance on benchmarks like TB Lite and TB 2.1 compared to its base model. It is part of a collection of specialized terminal agents.
Loading preview...
Overview
This model, allenai/qwen35-9b-endless, is a 9 billion parameter language model developed by Ai2. It is a specialized terminal-agent, fine-tuned from the Qwen 3.5 9B base model using Deep Proximal Policy Optimization (DPPO). The training specifically utilized the Endless Terminals dataset as an ablation study, making it highly proficient in command-line interface (CLI) environments.
Key Capabilities and Performance
- Terminal Agent Specialization: Optimized for interacting with and operating within terminal environments.
- Improved Benchmark Scores: Demonstrates enhanced performance over the base Qwen 3.5 9B model on terminal-specific benchmarks:
- Achieves 52.6 \u00b1 1.4 on TB Lite, up from 41.9 \u00b1 2.7.
- Achieves 25.5 \u00b1 1.4 on TB 2.1, up from 16.1 \u00b1 3.7.
- Training Details: Trained with DPPO using a max prompt token length of 2048 and a max overall token length of 65536, over 500 steps (with the best checkpoint at 200 steps for TBLite performance).
Use Cases
This model is particularly well-suited for:
- Automated Terminal Tasks: Developing agents that can autonomously navigate and execute commands in CLI environments.
- Research in Agentic AI: Exploring and building upon terminal-based agent capabilities.
- Benchmarking: Serving as a strong baseline or component for evaluating new terminal agent methodologies.