Downloads · 30 days
0
hongli-zhan/empathy-tactic-taggers-llama3.1-8b
empathy-tactic-taggers-llama3.1-8b is a text generation model from hongli-zhan. Use it when you need the model to write or continue text. It is set up for peft. The card lists the license as apache-2.0.
This repository contains 10 LoRA adapters built on top of meta-llama/Llama-3.1-8B-Instruct for sentence-level empathy tactic tagging.
Downloads · 30 days
0
Access
Public
Updated Apr 28, 2026
Repo size
6.7 GB
Likes
1
Public
Click a slice to open those files.
.safetensors6.7 GB · 97%
From the Hugging Face model README
This repository contains 10 LoRA adapters built on top of meta-llama/Llama-3.1-8B-Instruct for sentence-level empathy tactic tagging.
Each adapter predicts whether a target sentence expresses one specific empathy tactic, conditioned on the full response as context. These taggers implement the empathy tactic taxonomy used in Discourse Diversity in Multi-Turn Empathic Dialogue and are intended for research use in discourse analysis, tactic annotation, reward design, and evaluation of supportive dialogue.
| Tactic | Description |
|---|---|
| advice | Actionable suggestions or coping strategies |
| assistance | Offering personal help or support |
| emotional_expression | Communicating the supporter’s own feelings or reactions |
| empowerment | Affirming the seeker’s strength, worth, or capability |
| information | Providing facts, resources, or informational support |
| paraphrasing | Restating the seeker’s situation or feelings |
| questioning | Asking questions to better understand the seeker |
| reappraisal | Reframing the situation to shift its emotional meaning |
| self_disclosure | Sharing a related personal experience |
| validation | Reassuring or normalizing the seeker’s feelings |
On the held-out human-annotated test set reported in the paper, the taggers achieve an average macro F1 of 0.80 across 10 tactics.
Selected per-tactic F1 scores:
<score>0</score> or <score>1</score>python launch_tactic_tagger_server.py --port 8100
Each adapter is exposed as a separate model name such as advice or validation.
import requests
response = requests.post("http://localhost:8100/v1/chat/completions", json={
"model": "advice",
"messages": [
{"role": "system", "content": "You are a Fair Tagger Assistant..."},
{"role": "user", "content": "<prompt with full response + sentence>"},
],
"temperature": 0,
"max_tokens": 32,
})
content = response.json()["choices"][0]["message"]["content"]
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base_model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
model = PeftModel.from_pretrained(
base_model,
"hongli-zhan/empathy-tactic-taggers-llama3.1-8b",
subfolder="advice",
)
These adapters are research tools for empathy tactic analysis. They are not general-purpose emotion classifiers, dialogue-act taggers, mental health assessment tools, or safety classifiers.
@article{zhan2026discourse,
title={Discourse Diversity in Multi-Turn Empathic Dialogue},
author={Zhan, Hongli and Gueorguieva, Emma S and Hernandez, Javier and Suh, Jina and Ong, Desmond C and Li, Junyi Jessy},
journal={arXiv preprint arXiv:2604.11742},
year={2026}
}