Downloads · 30 days
74
11% of all-time downloads
Hyukkyu/Llama-3.1-8B-RAQUEL-MUSE-M-ret-LoRA-v1
Llama-3.1-8B-RAQUEL-MUSE-M-ret-LoRA-v1 is a text generation model from Hyukkyu. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as llama3.1.
Built with Llama. See LICENSE, NOTICE, and USEPOLICY.md.
Downloads · 30 days
74
11% of all-time downloads
All-time downloads
684
Public
Parameters
8B
33.5 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors16.7 GB · 100%
From the Hugging Face model README
Built with Llama. See LICENSE, NOTICE, and USE_POLICY.md.
A LoRA-trained M_ret baseline for the RAQUEL MUSE experiments. M_ret is trained from the pinned pretrained
meta-llama/Llama-3.1-8B on the retain source questions only of the RAQUEL2 muse_news source QA, question-answer pairs only
(no documents). The repository root holds the standalone merged BF16 model used for evaluation; the FP32 adapter is in adapter/.
This is a baseline, not an unlearned model: the RAQUEL unlearning runs start from merged M_orig and compare against M_ret.
Release note. This revision supersedes the earlier article-trained v1 checkpoint (still available at revision ae72ad45f8318161a7d78df8d6c5e9797eaa04d5). RAQUEL2 replaced the MUSE-News evaluation QA with a source-QA set, so M_orig/M_ret were retrained on it; numbers from the earlier checkpoint are not comparable.
| Split | M_ret | M_orig |
|---|---|---|
| Forget (original) | 176/750 (23.5%) | 99.7% |
| Forget (paraphrased) | 159/741 (21.5%) | 91.1% |
| Retain (original) | 1300/1300 (100.0%) | 100.0% |
| Retain (paraphrased) | 1141/1283 (88.9%) | 87.5% |
| RAQUEL affected | 153/1709 (9.0%) | 11.8% |
| RAQUEL unaffected | 256/2010 (12.7%) | 16.0% |
Semantic accuracy judged by Qwen/Qwen3.8-27B (vLLM 0.28.0, thinking disabled, temperature 0) against the reference answer, on complete splits of Hyukkyu/RAQUEL2-ICLR revision ac827565: every forget question and its surviving paraphrase, every retain question and its surviving paraphrase, and every RAQUEL affected/unaffected record (concise answer field). Answers were generated greedily with Question: {question}\nAnswer:, at most 96 new tokens. Per-split counts and evidence hashes are in evaluation.json.
meta-llama/Llama-3.1-8B, revision d04e592bb4f6aa9cfee91e2e20afa771667e1d4b.Hyukkyu/RAQUEL2-ICLR revision ac82756570fcce84441fb413ab523de8da679efd, config source-qa (1300 QA pairs).Question: {question}\nAnswer:.Exact settings and weight hashes are in training_recipe.json.
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
repo_id = "Hyukkyu/Llama-3.1-8B-RAQUEL-MUSE-M-ret-LoRA-v1"
tokenizer = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForCausalLM.from_pretrained(repo_id, dtype=torch.bfloat16, device_map="auto").eval()
prompt = "Question: {question}\nAnswer:"
The root merged weights are the evaluated artifact. The FP32 LoRA adapter is in adapter/
(PeftModel.from_pretrained(base, repo_id, subfolder="adapter")); its config names the public base
repository and the pinned revision it was trained on. Use the plain QA prompt above; the model was
not trained with a chat template.