EmPO: Emotion Grounding for Empathetic Response Generation through Preference Optimization
Paper • 2406.19071 • Published • 1
Mistral-7B — EmpatheticDialogues Fine-Tune
Repo Yapısı
| Klasör | İçerik |
|---|---|
final/ |
Model ağırlıkları (LoRA adapter) |
checkpoints/epoch_N/ |
Epoch checkpointları |
eval/ |
BLEU / ROUGE / BERTScore sonuçları |
logs/ |
Eğitim log |
Literatür Temeli
| Karar | Kaynak |
|---|---|
| Output < 20 kelime filtrele | Long Is More (arXiv:2402.04833) |
| Multi-turn context | SoulChat (arXiv:2311.00273) |
| lora_dropout=0.1 | LoRA Dropout (arXiv:2404.09610) |
| weight_decay=0.01 | AdaLoRA (arXiv:2303.10512) |
| Emotion etiketi + sistem promptu | EmPO (arXiv:2406.19071) |
Eğitim Sonuçları
| Training Loss | 1.1425 |
| Süre | 130.5 dk |
| Train samples | 3084 |
Yükleme
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
base = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2", torch_dtype=torch.float16, device_map="auto")
model = PeftModel.from_pretrained(base, "nvlz/mistral-ed-only", subfolder="final")
tokenizer = AutoTokenizer.from_pretrained("nvlz/mistral-ed-only", subfolder="final")
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for nvlz/mistral-ed-only
Base model
mistralai/Mistral-7B-Instruct-v0.2Papers for nvlz/mistral-ed-only
LoRA Dropout as a Sparsity Regularizer for Overfitting Control
Paper • 2404.09610 • Published
Long Is More for Alignment: A Simple but Tough-to-Beat Baseline for Instruction Fine-Tuning
Paper • 2402.04833 • Published • 5
SoulChat: Improving LLMs' Empathy, Listening, and Comfort Abilities through Fine-tuning with Multi-turn Empathy Conversations
Paper • 2311.00273 • Published
Adaptive Budget Allocation for Parameter-Efficient Fine-Tuning
Paper • 2303.10512 • Published • 4