edwixx/qwen3-8b-triton

TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 29, 2025License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The edwixx/qwen3-8b-triton model is an 8.19 billion parameter language model, fine-tuned from Alibaba Cloud's Qwen3-8B using a Triton-based pipeline. This model enhances the base Qwen3-8B's strong reasoning and instruction-following capabilities with task-specific adaptations via custom Triton kernels. It supports an extended context of up to 40,960 tokens and includes multi-step tool calling functionality. This model is ideal for general-purpose text generation, complex instruction following, and applications requiring advanced reasoning with tool integration.

Loading preview...

Model Overview

The edwixx/qwen3-8b-triton model is a fine-tuned version of Alibaba Cloud's Qwen3-8B, developed by Anurag Kanade. This 8.19 billion parameter model leverages a Triton-based fine-tuning pipeline to introduce task-specific adaptations through custom Triton kernels, building upon the robust foundation of the Qwen3 architecture.

Key Capabilities

  • Enhanced Instruction Following: Retains and improves upon the strong instruction-following capabilities of the base Qwen3-8B model.
  • Extended Context Window: Supports a maximum context length of 40,960 tokens, enabling processing of longer inputs and generating more coherent, extended outputs.
  • Multi-step Tool Calling: Features built-in support for tool/function calling, integrated via its chat template, facilitating complex interactive applications.
  • Advanced Reasoning: Incorporates support for explicit thinking/reasoning blocks (<think>...</think>) during generation, aiding in structured problem-solving.
  • General Text Generation: Proficient in general-purpose language understanding and generation tasks.

Technical Specifications

  • Base Model: Qwen/Qwen3-8B
  • Parameters: 8.19B (BF16 precision)
  • Architecture: Qwen3ForCausalLM with 36 layers and 32 attention heads (8 KV heads).
  • Attention Mechanism: RoPE (Rotary Position Embeddings) with a high theta of 1,000,000.

Ideal Use Cases

This model is well-suited for developers and researchers looking for a powerful 8B-class model that excels in:

  • Applications requiring complex instruction adherence and multi-turn conversations.
  • Scenarios benefiting from extended context processing.
  • Building agents or systems that need tool integration and structured reasoning.
  • General text generation and completion tasks where high fidelity and logical coherence are crucial.

Finetunes
1 models