Adiuk/eyla-qwen3-8b-tools-v2

TEXT GENERATIONPricing:Input $0.468 / Output $1.82Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 12, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Adiuk/eyla-qwen3-8b-tools-v2 is an on-device tool-calling model, a supervised fine-tune of Qwen3-8B, developed by Adioris / Eyla. It is designed to emit structured tool calls from natural-language intent in English, Bangla, and Banglish, without requiring tool schemas in context. This compact model is optimized for local execution, making it suitable for offline, privacy-preserving agentic workflows, particularly for Bangla-first products.

Loading preview...

Overview

Adiuk/eyla-qwen3-8b-tools-v2 is a specialized, on-device tool-calling model developed by Adioris / Eyla, built upon the Qwen3-8B architecture. It is specifically fine-tuned to interpret natural language requests and generate structured tool calls, supporting English, Bangla (বাংলা), and code-switched Banglish. A key differentiator is its ability to operate without requiring tool schemas in the inference context, as it has been trained to "compile tools & procedures into its weights."

Key Capabilities

  • On-device Tool-Calling: Designed for local execution, enabling offline and privacy-preserving AI assistant workflows.
  • Multilingual Support: Processes tool-calling requests in English, Bangla, and Banglish.
  • Schema-Free Inference: Infers tool calls directly from intent without needing tool schemas in the prompt, simplifying agentic system design.
  • Compact & Fast: Optimized for small footprint (4.3 GB for 4-bit quantization) and fast inference (5 seconds per turn on Apple Silicon), making it suitable for consumer hardware.
  • Structured Output: Emits tool calls in a consistent JSON-like format, e.g., <tool_call>{"name": "file_read", "arguments": {"path": "/path/to/file"}}</tool_call>.

Intended Use Cases

  • Agent Brain/Tool-Router: Functions as the core decision-maker within an agent harness for parsing and executing tool calls (e.g., file operations, web search, shell commands).
  • Offline Assistants: Ideal for scenarios where cloud data transmission is not feasible or desired.
  • Bangla-First Products: Specifically caters to applications requiring tool-calling capabilities from Bengali or Banglish user requests.

Finetunes
1 models