posttrainllm/qwen3-4b-file-ops-distilled

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 3, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The posttrainllm/qwen3-4b-file-ops-distilled model is a 4 billion parameter Qwen3-Instruct derivative, specifically distilled for multi-turn file-operation tasks within the GorillaFileSystem. Developed by posttrainllm, this specialist model achieves 100% on file-ops hard gate tasks, demonstrating superior performance in its niche compared to the stock Qwen3-4B. It is designed to be used behind a router for tasks involving file system navigation, creation, deletion, and manipulation.

Loading preview...

Qwen3-4B File-Ops Distilled: A Specialist LLM

This model, developed by posttrainllm, is a specialized 4 billion parameter Qwen3-Instruct derivative fine-tuned for multi-turn file-operation tasks. It is specifically designed for use with the GorillaFileSystem and functions as a routed specialist, not a general-purpose planner.

Key Capabilities & Performance

  • Exceptional File Operations: Achieves 100% on "File-ops hard gate" and 95% on "File-ops hardgen held-out" benchmarks, significantly outperforming the stock Qwen3-4B (58% on hard gate).
  • Distilled Training: Utilizes frontier/gold trajectory distillation, rendered in the student's native tool-calling chat template, to optimize for file system interactions.
  • Precision: The model is in bf16 precision.

Recommended Use Cases

  • File System Automation: Ideal for tasks requiring precise file system navigation, creating, deleting, moving, and renaming files or directories.
  • Router Integration: Best used when a router has already identified a file-operation task with derivable arguments, ensuring the model is applied to its intended domain.

Limitations

  • Domain Specificity: Performance outside the file-operation domain shows negative transfer, with out-of-domain breadth dropping to 42.3% compared to the stock model's 59.6%. It is not suitable as a general planner.
  • Routing Dependency: Relies on correct routing; incorrect routing can lead to significant performance regression.

This model is distributed under the Apache-2.0 license, derived from Qwen/Qwen3-4B-Instruct-2507.

Finetunes
1 models