reaperdoesntknow/LFM2.5-1.2B-Distilled-SFT

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.2BQuant:BF16Context Size:32kPublished:Mar 25, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

LFM2.5-1.2B-Distilled-SFT is a 1.2 billion parameter hybrid SSM + attention model developed by Convergent Intelligence LLC: Research Division. It was created through knowledge distillation from a 24B MoE hybrid teacher on STEM chain-of-thought data, followed by supervised fine-tuning on logical inference. This model is optimized for structured STEM reasoning and formal logical inference, offering efficient on-device performance with sub-1GB RAM usage and high token throughput on AMD CPUs, making it suitable for edge deployments and formal reasoning tasks.

Loading preview...

LFM2.5-1.2B-Distilled-SFT: Efficient STEM Reasoning and Logical Inference

This 1.2 billion parameter model, developed by Convergent Intelligence LLC: Research Division, is a unique hybrid architecture combining State Space Models (SSM) with attention mechanisms. It's engineered for high efficiency and specialized reasoning capabilities, making it ideal for resource-constrained environments.

Key Capabilities

  • Hybrid Architecture: Leverages SSM components for efficient sequential state propagation, particularly beneficial for logical inference, combined with attention for broader context.
  • Distilled STEM Reasoning: Knowledge distilled from a 24B MoE hybrid teacher model using a proof-weighted cross-entropy and KL divergence loss, focusing on STEM chain-of-thought data across linear algebra, differential equations, electromagnetism, mathematics, and classical mechanics.
  • Logical Inference Specialization: Supervised fine-tuned on the LogicInference dataset, activating the hybrid architecture's natural alignment with propositional logic chains, where each premise updates a logical state.
  • High Efficiency: Designed for on-device deployment, achieving 239 tokens/second on AMD CPUs and fitting under 1GB of RAM, with a context length of 1024 tokens.

Good For

  • On-device logical inference and STEM reasoning: Ideal for mobile, edge, and IoT deployments requiring structured reasoning.
  • Formal reasoning tasks: Excels in scenarios demanding rigorous derivation and propositional logic.
  • Educational tutoring and embedded inference pipelines: Suitable for applications where efficient, specialized reasoning is crucial.
  • Resource-constrained environments: Provides strong reasoning capabilities within a small footprint.

Finetunes
1 models