modrill/qwen3-4b-think-baseline-lora-sft

TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 7, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

The modrill/qwen3-4b-think-baseline-lora-sft model is a 4 billion parameter Qwen3-based causal language model developed by modrill, fine-tuned using LoRA for enhanced performance. This model is specifically optimized for 'Think' mode, enabling advanced reasoning capabilities, and supports a context length of up to 32768 tokens. It is designed for direct inference with merged LoRA adapters, making it suitable for applications requiring sophisticated problem-solving and code generation in both English and Chinese.

Loading preview...

Qwen3-4B Think Baseline (LoRA SFT)

This model, developed by modrill, is a 4 billion parameter variant of the Qwen3-4B-Base model, fine-tuned using LoRA (rank 64, alpha 128) with its adapters merged into the full weights for streamlined inference. It is specifically designed to operate in 'Think' mode, which enhances its reasoning and problem-solving capabilities. The model supports a substantial training cutoff length of 24576 tokens, indicating its proficiency in handling complex and lengthy inputs.

Key Capabilities

  • Enhanced Reasoning: Optimized for 'Think' mode, suggesting improved logical processing and problem-solving.
  • Code Generation: Tagged for 'code' applications, indicating suitability for programming-related tasks.
  • Multilingual Support: Supports both English and Chinese languages.
  • Direct Inference: LoRA adapters are merged into the full weights, allowing for direct use without separate adapter loading.

Good For

  • Applications requiring advanced reasoning and 'thinking' processes.
  • Code generation and understanding tasks.
  • Projects needing a compact yet capable 4B parameter model with extended context handling.
  • Multilingual text generation in English and Chinese.

Finetunes
1 models