TeichAI/LFM2.5-1.2B-Thinking-Pony-Alpha-Distill

TEXT GENERATIONConcurrency Cost:1Model Size:1.2BQuant:BF16Ctx Length:32kPublished:Feb 11, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

TeichAI/LFM2.5-1.2B-Thinking-Pony-Alpha-Distill is a 1.2 billion parameter language model developed by TeichAI, built upon the LiquidAI/LFM2.5-1.2B-Thinking base model. It is specifically fine-tuned on a large reasoning dataset derived from Pony Alpha, an early checkpoint of GLM-5. This model is optimized for complex reasoning tasks, making it suitable for applications in coding, science, and deep research.

Loading preview...

Model Overview

TeichAI/LFM2.5-1.2B-Thinking-Pony-Alpha-Distill is a 1.2 billion parameter language model, leveraging the LiquidAI/LFM2.5-1.2B-Thinking as its base. This model has been specifically trained on a substantial reasoning dataset, TeichAI/Pony-Alpha-15k, which was derived from Pony Alpha, an early checkpoint of the GLM-5 model. The training process for this LFM-2.5 model was accelerated by 2x using Unsloth and Hugging Face's TRL library.

Key Capabilities

  • Enhanced Reasoning: Fine-tuned on a dedicated reasoning dataset for improved logical processing.
  • Efficient Training: Utilizes Unsloth and Hugging Face TRL for faster model development.

Good For

This model is particularly well-suited for applications requiring strong analytical and problem-solving abilities:

  • Coding: Generating or assisting with programming tasks.
  • Science: Processing and understanding scientific texts and data.
  • Deep Research: Aiding in complex information retrieval and analysis.

Recommended Sampling Parameters

For optimal performance, Liquid AI suggests the following sampling parameters:

  • temperature: 0.05
  • top_k: 50
  • repeat_penalty: 1.05

Finetunes
2 models