metadata

license: llama3
datasets: Microsoft/ChatBench
language: en
base_model: meta-llama/Meta-Llama-3-8B
library_name: peft
tags:
  - Microsoft
  - ChatBench
  - Interactive Benchmark
  - User Simulator
  - Benchmarking

chatbench-llama3-8b

Overview

ChatBench Simulators are fine-tuned user simulators designed to enable automated, realistic evaluation of large language models (LLMs) through simulated user–AI conversations.

Instead of recruiting human participants for every evaluation, you can use this simulator (chatbench-llama3-8b) to act as a proxy user. By providing a multiple-choice question (full stem plus answer options) via a CLI or Python API, the simulator generates natural user turns—asking clarifications, signaling understanding, or indicating errors—until the simulated user “accepts” an answer.

These conversations can be used to compute task success rate, coherence, user satisfaction, error recovery, and latency metrics, enabling researchers and practitioners to move beyond static benchmarks and evaluate models under interactive, user-in-the-loop conditions.

GitHub | Paper

Model Details

Model Description

Model type: Causal LM (user simulator, LoRA adapter on Llama-3-8B)
Base model: meta-llama/Meta-Llama-3-8B
Fine-tuned on: microsoft/ChatBench (≈12K multi-turn QA dialogs)
Languages: English
License: Inherits from Meta-Llama-3 (llama3)
Developed by: Microsoft
Contacts: [email protected], [email protected], [email protected]

Training Setup:

The model was fine-tuned end-to-end on ChatBench data using the same simulator recipe described in Section 5 of the paper. Each example is formatted as:

[SYSTEM] <instruction>

<previous turns>

[USER] 
→ model generates: <assistant reply> [END]

Optimizer: AdamW
Adapter: LoRA (r=8, α=32, dropout=0.05)
Quantization: 8-bit weights with CPU offload
Precision: bf16 (trained on 4× RTX A6000 GPUs)
Quantization: 8-bit CPU offload for inference
Batch size: 1 per GPU
Epochs: 2
LR: 5e−5

Intended Uses

Direct Use

Automated user simulation for interactive benchmarking of LLMs.
Research reproduction of the ACL’25 paper results.
Prototyping multi-turn evaluation pipelines with a mid-size simulator model.

Out-of-Scope

Deployment in sensitive domains (healthcare, legal, financial) without expert oversight.
Open-ended creative generation (storytelling, brainstorming).
Very long-context tasks beyond the Llama-3-8B context window.

Bias, Risks, and Limitations

Bias: Inherits biases from Llama-3-8B pretraining and ChatBench data.
Factual reliability: May hallucinate or produce unrealistic user behaviors outside the training domain.
Coverage: Optimized for structured, multiple-choice QA; not suitable for unconstrained dialogue.
Adapter trade-offs: LoRA adapters may underfit compared to full fine-tunes.

How to Get Started

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# Load base Llama-3-8B in 8-bit
base = AutoModelForCausalLM.from_pretrained(
    "meta-llama/Meta-Llama-3-8B",
    load_in_8bit=True,
    device_map="auto"
)

# Load ChatBench LoRA adapter
model = PeftModel.from_pretrained(base, "microsoft/chatbench-llama3-8b")

tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B")
tokenizer.pad_token = tokenizer.eos_token

inputs = tokenizer(
    "[SYSTEM] You are a user.\n\n[USER] What is 2+2?\n\n[USER] ",
    return_tensors="pt"
).to("cuda")

outputs = model.generate(**inputs, max_new_tokens=64)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Evaluation

Perplexity

We report perplexity (PPL) on held-out ChatBench data, comparing baseline Llama3-8B vs. fine-tuned chatbench-llama3-8b.

A lower PPL means the fine-tuned model is closer (more confident) in the human-like reply it was trained on.

Model	Perplexity
Llama-3-8B (baseline)	4.14
chatbench-llama3-8b	1.89

Interactive Evaluation (ACL’25 Study)

Correlation with Human User–AI Accuracy: +20 point improvement over unfine-tuned baselines.
Task Success Accuracy: Within ±5 points of real user–AI results across five MMLU subsets.
Ablations: Removing persona-conditioning or chain-of-thought prompts reduced coherence scores by ~10 points.

Full details are available in Section 6 of the paper.

Technical Specifications

Compute Infrastructure

Hardware: 4× NVIDIA RTX A6000 GPUs (48GB VRAM each), 128-core x86_64 CPU
Software: Ubuntu 22.04, CUDA 12.4, PyTorch + Hugging Face Transformers + PEFT

Citation

https://arxiv.org/abs/2504.07114

Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computers and Society (cs.CY); Human-Computer Interaction (cs.HC)

Cite as:

arXiv:2504.07114 [cs.CL] (or arXiv:2504.07114v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2504.07114

BibTeX:

@misc{chang2025chatbenchstaticbenchmarkshumanai, title={ChatBench: From Static Benchmarks to Human-AI Evaluation}, author={Serina Chang and Ashton Anderson and Jake M. Hofman}, year={2025}, eprint={2504.07114}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2504.07114}, }

APA:

Chang, S., Anderson, A., & Hofman, J. M. (2025). ChatBench: From Static Benchmarks to Human-AI Evaluation. arXiv [Cs.CL]. Retrieved from http://arxiv.org/abs/2504.07114