notnotsamuel/LFM2.5-350M-RLCD

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.35BQuant:BF16Context Size:32kPublished:Sep 16, 2026License:otherArchitecture:Transformer0.0K Featherless Exclusive Cold

The notnotsamuel/LFM2.5-350M-RLCD model is a 0.35 billion parameter language model based on LiquidAI/LFM2.5-350M, specifically designed for parallel structured inference. It reuses the original model weights but implements a novel constrained generation method to programmatically assemble JSON outputs. This approach significantly accelerates structured data extraction, achieving up to 62.91x speedups compared to autoregressive generation for tasks like extracting boolean and string enum fields from text. It excels at guaranteeing valid and schema-compliant JSON output for specific data extraction use cases.

Loading preview...

Overview

notnotsamuel/LFM2.5-350M-RLCD is a 0.35 billion parameter model that leverages the original LiquidAI/LFM2.5-350M weights for parallel structured inference. This model focuses on efficiently generating valid JSON outputs by programmatically assembling results, rather than relying on the model to generate the structure autoregressively. It's designed for inference only, with no training or fine-tuning involved.

Key Capabilities

  • Accelerated Structured Inference: Achieves significant speedups (e.g., 8.46x on M2 Max, 62.91x on L40S) for structured data extraction compared to traditional autoregressive generation.
  • Guaranteed JSON Validity: Ensures 100% valid and schema-compliant JSON output for supported schemas by assembling the JSON programmatically.
  • Schema-Driven Extraction: Supports flat JSON objects with required boolean or string enum properties, allowing users to define extraction tasks via JSON Schema.
  • Reuses Base Model Weights: Utilizes the unchanged weights of LiquidAI/LFM2.5-350M, ensuring reproducibility and consistency with the base model's capabilities.

Good For

  • Structured Data Extraction: Ideal for use cases requiring the extraction of specific, predefined fields (booleans, enum strings) into a JSON format from unstructured text.
  • Reducing Latency for JSON Output: Suitable for applications where fast, reliable JSON output is critical, such as classifying customer support messages or extracting attributes.
  • Ensuring Output Format Compliance: When strict adherence to a JSON schema is paramount, this method guarantees structural validity, even if the model's 'decision quality' for individual fields may vary.

Finetunes
1 models