shuhulx/Qwopus3.5-4B-Coder-Fable5-v1
shuhulx/Qwopus3.5-4B-Coder-Fable5-v1 is a 4.5 billion parameter Qwen3.5-based causal language model, fine-tuned as a continuation of Jackrong/Qwopus3.5-4B-Coder. It is specifically trained on Glint-Research/Fable-5-traces, a dataset of Claude Fable 5 local coding-agent trajectories, emphasizing tool-use, debugging, and agentic reasoning. This model excels at agentic coding workflows, including repository inspection, tool calling (Bash, Read, Write, Edit), and debugging loops for local development environments.
Loading preview...
Qwopus3.5-4B-Coder-Fable5-v1: Agentic Coding Model
This model is a 4.5 billion parameter continuation of the Jackrong/Qwopus3.5-4B-Coder base, built upon the Qwen3.5 architecture. Its primary differentiator is specialized training on the Glint-Research/Fable-5-traces dataset, which comprises Claude Fable 5 local coding-agent trajectories. This dataset focuses on multi-step agent workflows, including tool-use, local command context, code editing, debugging loops, and <think>-style reasoning completions, making it distinct from general-purpose code models.
Key Capabilities
- Agentic Coding: Designed for iterative coding-agent loops involving repository inspection, planning, tool invocation, file modification, and validation.
- Tool-use Workflows: Optimized for structured tool interactions such as Bash commands, file reading, writing, and editing, crucial for developer automation.
- Debugging & Repair: Proficient in identifying failing files, interpreting stack traces, planning test commands, proposing minimal patches, and iterating through error resolution.
- Trace-style Reasoning: Handles long-form planning and explicit reasoning traces, enabling more complex problem-solving.
When to Use This Model
This model is ideal for local-first deployment scenarios where a compact, specialized agent is needed for:
- Automating development tasks within a local environment.
- Implementing Hermes-style, Claude-Code-style, or OpenCode-style coding agents.
- Debugging codebases, analyzing errors, and suggesting fixes.
- Generating structured tool-use outputs for integration with external systems.
It is available in various formats, including Transformers, GGUF, MLX, and MLX 4-bit, supporting diverse inference environments like Python, llama.cpp, LM Studio, and Apple Silicon workflows.