sag-uniroma2/FrameLLaMA-3.1-8B-Instruct-FullFN17

TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Apr 15, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

FrameLLaMA-3.1-8B-Instruct-FullFN17 is an 8 billion parameter instruction-tuned causal language model developed by sag-uniroma2, fine-tuned from Llama-3.1-8B-Instruct. It injects structured knowledge from FrameNet 1.7 using parameter-efficient LoRA fine-tuning, focusing on principle-oriented supervision. This model excels at event-level semantic reasoning, improving tasks like Natural Language Inference (NLI) and Semantic Role Labeling (SRL) by understanding event structure and participant roles.

Loading preview...

Overview

FrameLLaMA-3.1-8B-Instruct-FullFN17 is an 8 billion parameter instruction-tuned causal language model, fine-tuned by sag-uniroma2 from Llama-3.1-8B-Instruct using LoRA. Its core innovation lies in integrating structured knowledge from FrameNet 1.7 through "principle-oriented supervision," where frame definitions, participant roles, and semantic relations are converted into structured question-answer tasks. This approach enables the model to learn reusable semantic constraints rather than isolated facts, enhancing its ability to handle event-level semantic reasoning.

Key Features

  • Full FrameNet Coverage: Trained on over 1,200 frames from FrameNet 1.7.
  • Principle-Oriented Learning: Encodes role constraints, semantic types, and frame relations.
  • Event-Level Reasoning: Significantly improves understanding of causality, entailment, and contradiction.
  • Frame-Aware Inference: Better handles lexical ambiguity and role compatibility.
  • Parameter-Efficient Training: Utilizes LoRA for scalable adaptation.

Performance & Use Cases

The model shows strong gains in entailment and contradiction detection on SNLI diagnostic subsets, and improved frame identification and role-span alignment in CONLL-style FrameNet SRL datasets. It reduces reliance on surface-level lexical cues, making it suitable for:

  • Natural Language Inference (NLI): Especially for event-based reasoning.
  • Semantic Role Labeling (SRL): For accurate frame and role prediction.
  • Event Understanding: Modeling causality and participant structure.
  • Linguistically-Informed AI: Applications requiring structured semantic interpretation.

Finetunes
1 models