DireDreadlord/Gallium-350M

TEXT GENERATIONConcurrent Unit Cost:1Model Size:0.35BQuant:BF16Context Size:32kPublished:Jul 8, 2026License:lfm1.0Architecture:Transformer0.0K Featherless Exclusive Cold

DireDreadlord/Gallium-350M is a 350 million parameter code reasoning and generation model built upon the LFM2.5 base architecture. It is specifically fine-tuned for accurate and quick code snippet and long-form code generation across major programming languages. This model excels in code reasoning, instruction following, and subject matter expert capabilities on code-related subjects, making it suitable for deployment on laptop-grade GPUs.

Loading preview...

Gallium-350M: Lightweight Code Reasoning and Generation

Gallium-350M is a 350 million parameter model developed by DireDreadlord, based on the LFM2.5 architecture. It is designed for efficient code reasoning and generation, offering accurate and rapid production of both code snippets and long-form code in various programming languages. Its small size allows for comfortable operation on laptop-grade GPUs.

Key Capabilities

  • Code Generation: Provides accurate and quick code snippet and long-form code generation.
  • Enhanced Reasoning: SFT and RL training on code reasoning traces (from the deepseek-v4-reasoning-code-2500 dataset) imbues the model with advanced reasoning capabilities, leading to higher quality and hallucination-free outputs.
  • Instruction Following: Capable of following instructions for code-related tasks.
  • Subject Matter Expertise: Offers Q/A and subject matter expert capabilities on code-related subjects.
  • Efficiency: A significant upgrade from previous GemCod models with minor bloat to inference time and space requirements.

Intended Use Cases

  • Generating code snippets from natural language prompts.
  • Creating long-form code based on detailed instructions.
  • Reasoning through complex coding problems.
  • Answering questions and providing expertise on code-related topics.