violetxi/qwen35-9b-harvey-v4-notes-conditioned-1m
The violetxi/qwen35-9b-harvey-v4-notes-conditioned-1m is a 9 billion parameter Qwen3.5-based language model, fine-tuned for agentic behavior using a unique dataset of 1 million notes and note-conditioned trajectories. It features a 32768 token context length and is specifically optimized for tasks requiring historical thinking and agentic reasoning. This model is designed for applications where an AI agent needs to process and act upon structured notes and complex interaction histories.
Loading preview...
Model Overview
This model, violetxi/qwen35-9b-harvey-v4-notes-conditioned-1m, is a 9 billion parameter variant of the Qwen3.5 architecture. It has been specifically fine-tuned using a unique dataset comprising 1 million notes and note-conditioned trajectories, with a 70% / 30% supervised-token mixture. The training involved two epochs and 60 optimizer updates, building upon the Qwen/Qwen3.5-9B base model.
Key Training Details
- Base Model:
Qwen/Qwen3.5-9B - Dataset: 1 million supervised tokens from notes and note-conditioned trajectories.
- Training Objective: Causal next-token prediction for notes and trajectory loss for assistant labels.
- Context Length: Utilizes a packed sequence length of 16,384 tokens during training, implying a high effective context window.
Evaluation and Performance
The model's performance was evaluated on historical thinking-enabled agent generations, regraded with gpt-5.6-sol using the Harvey per-criterion rubric. It achieved an all-criteria-pass rate of 2.30% with a 5-turn budget and 3.10% with a 20-turn budget on specific evaluation datasets. This evaluation methodology focuses on the model's ability to pass every criterion in complex agentic tasks, indicating its specialized capability in structured reasoning and interaction.
Intended Use Cases
This model is particularly suited for applications requiring advanced agentic capabilities, especially those involving:
- Historical Thinking: Tasks that benefit from processing and reasoning over past notes and interaction histories.
- Note-Conditioned Trajectories: Scenarios where an agent's actions are guided or informed by structured notes.
- Complex Agentic Systems: Development of AI agents that need to perform multi-step reasoning and decision-making based on detailed contextual information.