manishkumar2101114/qwen3.6-35b-a3b-tool-merged-sglang

TEXT GENERATIONPricing:Input $0.4 / Cached $0.07 / Output $4Concurrent Unit Cost:2Model Size:35.1BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 10, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The manishkumar2101114/qwen3.6-35b-a3b-tool-merged-sglang model is a 35.1 billion parameter language model based on the Qwen3.6-35B-A3B architecture. It is specifically fine-tuned for tool-calling capabilities, utilizing a LoRA adapter trained on 1515 tool-call dialogues. This model is designed for applications requiring robust function calling and integration with external tools, offering enhanced performance in structured interaction scenarios.

Loading preview...

Overview

This model, manishkumar2101114/qwen3.6-35b-a3b-tool-merged-sglang, is a specialized variant of the Qwen3.6-35B-A3B architecture, featuring 35.1 billion parameters and a 32768-token context length. It is built upon the Qwen/Qwen3.6-35B-A3B-FP8 base model, with FP8 weights dequantized to BF16 during the merging process. The key differentiator is its fine-tuning with a LoRA adapter specifically for tool-calling, making it highly proficient in interacting with external functions and APIs.

Key Capabilities

  • Advanced Tool Calling: Optimized for robust and accurate function calling, enabling seamless integration with external tools and services.
  • Qwen3.6-35B-A3B Base: Leverages the strong foundational capabilities of the Qwen3.6-35B-A3B model.
  • Efficient Fine-tuning: Utilizes a LoRA adapter (r=32) trained over 3 epochs on 1515 tool-call dialogues, demonstrating effective specialization with a low training loss of 0.2029 and evaluation loss of 0.2047.
  • BF16 Precision: Operates with BF16 precision, balancing performance and computational efficiency.

Good For

  • Agentic Workflows: Developing AI agents that need to interact with external systems and execute specific functions.
  • Complex Task Automation: Automating multi-step processes that require calling various tools or APIs.
  • Structured Interactions: Applications where the model needs to generate structured outputs for tool invocation rather than just free-form text.
  • Developers requiring a Qwen-based tool-calling model: Offers a specialized solution within the Qwen ecosystem for tool-use cases.

Finetunes
1 models