Skip to content
Other

FrameLLaMA-3.1-8B-Instruct-50FramesFN17

A model for frame semantics analysis

2 min read

This language model is built on Llama-3.1-8B-Instruct and specializes in understanding FrameNet frames (definitions, frame element roles, and lexical units). It helps research text semantic structure and perform semantic role labeling tasks without prior training on specific SRL datasets. The new release is useful for NLP researchers, linguists, and developers of semantic search engines.

Key features

  • FrameNet specialization: Understanding of 60 frames and their elements.
  • High accuracy: Outperforms the base model on closed and open-ended questions.
  • Efficient adaptation: Uses LoRA (a small add-on to the base model) with a rank of 16.
  • Scientific background: Presented at the *SEM 2025 conference.

What you can do

  • Explore semantics: Ask questions about definitions and roles of specific frames.
  • Test SRL: Perform zero-shot semantic role labeling in text.
  • Integrate into pipelines: Use for sentence semantic parsing in NLP projects.

Example

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

MODEL_ID = "sag-uniroma2/FrameLLaMA-3.1-8B-Instruct-50FramesFN17"
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForCausalLM.from_pretrained(MODEL_ID, torch_dtype=torch.bfloat16, device_map="auto")

question = "Can you list some frame elements in the Supply frame?"
prompt = f"### Input: {question}\n### Response:"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
output = model.generate(**inputs, max_new_tokens=64)
print(tokenizer.decode(output[0]))
Read moreDetails
Benchmarks: closed questions (F1 0.93), multiple choice (F1 0.66).
Training: 7 epochs, learning rate 2e-4, batch size 4.
Limitations: trained on only 60 FrameNet frames, incomplete coverage for roles is possible.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 15 GB · ready files17 GB
  • FP8 / INT8recommendedweights 7.5 GB8.7 GB
  • 4-bitweights 4.2 GB5.1 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 3080 · BF16 — from RTX 4090

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # sag-uniroma2/FrameLLaMA-3.1-8B-Instruct-50FramesFN17
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download sag-uniroma2/FrameLLaMA-3.1-8B-Instruct-50FramesFN17 --local-dir ~/models/FrameLLaMA-3.1-8B-Instruct-50FramesFN17
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

  1. Meta-Llama-3.1-8Bfamily root
  2. Llama-3.1-8B-Instruct
  3. FrameLLaMA-3.1-8B-Instruct-50FramesFN17

New posts about versions of Meta-Llama-3.1-8B (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

sag-uniroma2/FrameLLaMA-3.1-8B-Instruct-50FramesFN17
Author
sag-uniroma2
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments