Jepoxy/LFM2.5-350M-Memory-Extractor

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.35BQuant:BF16Context Size:32kPublished:Sep 2, 2026Architecture:Transformer0.0K Featherless Exclusive Cold

Jepoxy/LFM2.5-350M-Memory-Extractor is a 350 million parameter language model, fine-tuned from LFM2.5-350M-Base by Jepoxy. This lightweight model specializes in extracting concise, durable memories from conversational text, distilling transient information into persistent memory structures. It is designed for experimental memory extraction pipelines where efficiency is crucial, offering a specialized solution for converting dialogue into structured memory. The model was trained on 64K synthetic examples to achieve its focused memory extraction capability.

Loading preview...

Overview

Jepoxy/LFM2.5-350M-Memory-Extractor is a specialized 350 million parameter model, fine-tuned from the LFM2.5-350M-Base architecture. Its core function is to transform conversational text into structured, "durable memories." This model was developed by Jepoxy through training on 64,000 synthetic examples, where a teacher model identified and extracted key information from prompts that would be valuable to remember over time.

Key Capabilities

  • Memory Extraction: Distills transient conversational content into persistent, structured memory formats.
  • Lightweight Design: A small model (350M parameters) intended for efficient memory extraction, particularly in scenarios where larger models would be cost-prohibitive.
  • Structured Output: Generates JSON output containing extracted memories and a confidence score.

Use Cases

  • Experimental Memory Pipelines: Ideal for exploring and prototyping memory systems for AI agents or conversational interfaces.
  • Cost-Efficient Processing: Suitable for applications requiring memory extraction without the computational overhead of larger language models.

Limitations

It's important to note that this model is an experimental proof of concept, not a production-ready system. Its performance and consistency may vary, especially with inputs outside the scope of its 64K synthetic training examples. It serves as a starting point for investigating small-model memory extraction rather than a fully robust solution.

Finetunes
1 models