ZhiyangQi97's picture
Document public model access
22a7efa verified
|
Raw History Blame Contribute Delete
4.67 kB
metadata
license: apache-2.0
language:
  - zh
library_name: transformers
pipeline_tag: text-generation
base_model: Qwen/Qwen2.5-7B-Instruct
base_model_relation: finetune
tags:
  - counseltraj
  - counseltrajlm
  - counseling
  - conversational
  - qwen2
  - merged

CounselTrajLM — Qwen2.5-7B

📄 Paper: arXiv:2611.24168

CounselTrajLM is a Chinese counseling dialogue model fine-tuned on CounselTrajD to generate the next counselor response from dialogue history. This version uses Qwen2.5-7B-Instruct as its base.

This repository contains the complete merged BF16 model in nine safetensors shards, together with its tokenizer, chat template, inference script, and checksums. It can be loaded directly with Transformers.

Model and training details

Item Value
Base model Qwen/Qwen2.5-7B-Instruct
Base revision a09a35458c702b33eeacc393d103063234e8bc28
Training data CounselTrajD; dialogue-disjoint 90/10 train/validation split
Supervised training examples 425,679
Validation examples 47,528
Fine-tuning method QLoRA; adapter merged into the unquantized BF16 base for release
LoRA rank / alpha / dropout 32 / 64 / 0.05
Learning rate 1e-5
Schedule / warmup ratio Cosine / 0.06
Epochs 3; final-epoch export
Effective batch size 16
Maximum training sequence length 2,048 tokens
Release format Merged BF16 safetensors

Validation was used for monitoring. The released model is the final epoch. See training_summary.json for the recorded settings.

Usage

The repository is public and can be downloaded without authentication. Install the dependencies in requirements.txt.

pip install -r requirements.txt
python generate_reply.py --messages examples/history.json

The script accepts a JSON list of chat messages with role and content keys. Use user for client messages and assistant for counselor messages, ending with a client message. It adds the training system instruction if none is supplied. The example dialogue is newly written for demonstrating inference.

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "UEC-InabaLab/CounselTrajLM-Qwen2.5-7B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id, dtype=torch.bfloat16, device_map="auto"
)
messages = [
    {"role": "system", "content": "你是一名专业、温和、自然的线上心理咨询师。请根据对话历史生成下一句咨询师回复。回复只写咨询师要说的话,不要写说话人名称,不要写对话标签。"},
    {"role": "user", "content": "最近工作任务很多,下班以后也一直在想工作的事,很难放松。"},
]
prompt = tokenizer.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
with torch.inference_mode():
    output = model.generate(
        **inputs, max_new_tokens=160, do_sample=True,
        temperature=0.7, top_p=0.8, top_k=20, repetition_penalty=1.05,
    )
print(tokenizer.decode(output[0, inputs.input_ids.shape[-1]:], skip_special_tokens=True))

The weights occupy approximately 15.2 GB, with additional memory required for inference. The release was checked with Transformers 5.13.0, PyTorch 2.5.1, Accelerate 1.13.0, and safetensors 0.8.0. Use SHA256SUMS and validation.json to check the release files and validation status.

Intended use and limitations

The model is intended for research on Chinese counseling dialogue generation. Its training data are synthetic. Generated responses may be inaccurate, inappropriate, or unsafe, and the model is not a substitute for professional mental health care.

The model is released under Apache-2.0; see LICENSE. The source data's licensing and attribution information is provided in the CounselTraj repository.

Citation

@inproceedings{qi-inaba-2027-counseltraj,
    title = "{CounselTraj}: Synthesizing Process-Aware Psychological Counseling Dialogues via Dialogue-Act Trajectory Planning",
    author = "Qi, Zhiyang and Inaba, Michimasa",
    booktitle = "Proceedings of the 20th Conference of the European Chapter of the Association for Computational Linguistics",
    year = "2027",
    url = "https://arxiv.org/abs/2611.24168",
}