davidnichols-ops/qwen3-4b-devin-sft

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 2, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The davidnichols-ops/qwen3-4b-devin-sft is a 4 billion parameter instruction-tuned causal language model based on Qwen3-4B. It is specifically fine-tuned on Devin session data to excel at tool-calling tasks, particularly for engineering assistance. This model is optimized for generating structured tool calls in a JSON format, making it suitable for automated task execution and interaction with external systems. It features a 32K context length and is designed for efficient inference, especially when paired with a draft model for speculative decoding.

Loading preview...

Model Overview

The davidnichols-ops/qwen3-4b-devin-sft is a 4 billion parameter instruction-tuned model built upon the Qwen3-4B base architecture. Its primary distinction lies in its specialized fine-tuning on Devin session data, enabling it to proficiently handle tool-calling tasks.

Key Capabilities & Features

  • Tool-Calling Specialization: Fine-tuned to understand and generate structured tool calls, expecting a specific JSON format within <tool>{json}</tool> tags.
  • Engineering Assistant: Designed to function as a helpful engineering assistant, capable of interpreting user requests and formulating appropriate tool commands.
  • Efficient Training: Trained on a powerful NVIDIA GB10 system with 128GB VRAM, utilizing 4,992 samples from the davidnichols-ops/adaptive-operator-v4-dataset over 3 epochs.
  • BF16 Precision: Utilizes BF16 precision for training and inference, balancing performance and memory usage.
  • Speculative Decoding Support: Can be paired with a smaller draft model, such as davidnichols-ops/qwen3-0.6b-devin-draft, to achieve 2-3x inference speedup through speculative decoding.

Use Cases

This model is particularly well-suited for applications requiring:

  • Automated Code Execution: Generating commands for shell or other tools to run tests, fix issues, or perform development tasks.
  • Intelligent Agents: Powering AI agents that need to interact with external systems or APIs via structured tool calls.
  • Developer Tools: Integrating into IDEs or development workflows to provide automated assistance and task execution.

Finetunes
2 models