DhruvalLabs/qwen3-8b-claude-agentic-fable5
DhruvalLabs/qwen3-8b-claude-agentic-fable5 is a Qwen3-8B model fine-tuned by DhruvalLabs for agentic coding tasks. This model is specifically trained on Claude Fable-5 agent traces to enable step-by-step reasoning within tags and accurate tool calling for multi-step coding tasks. It functions as an agentic coding assistant, capable of planning and executing complex tasks autonomously using tools like bash, read_file, write_file, edit_file, and web_search. The model's training on real Fable-5 coding sessions makes it adept at emulating that agentic style.
Loading preview...
Model Overview
This model, developed by DhruvalLabs, is a Qwen3-8B base model that has been fine-tuned to operate as an agentic coding assistant. Its core capability lies in its ability to reason step-by-step within <think> tags before executing actions, mimicking the behavior of Claude's Fable-5 agent.
Key Capabilities
- Agentic Reasoning: Employs a reasoning chain within
<think>blocks before performing any action. - Tool Utilization: Proficiently calls various tools such as
bash,read_file,write_file,edit_file, andweb_searchto complete tasks. - Multi-step Task Execution: Designed to plan and execute complex coding tasks autonomously over multiple steps.
- Fable-5 Agent Style: Trained on real coding agent trajectories from the Glint-Research/Fable-5-traces dataset, enabling it to follow the distinct Fable-5 agent methodology.
Training Details
The model was fine-tuned using QLoRA on a dataset of 4,665 rows of Fable-5 agent traces. The training involved 580 steps with a sequence length of 8192, targeting specific Qwen3-8B modules like q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, and down_proj.
Ideal Use Cases
This model is particularly well-suited for applications requiring an intelligent coding assistant that can:
- Automate complex coding workflows.
- Perform code analysis and modification through tool interactions.
- Execute multi-step development tasks with autonomous reasoning.