Instructions to use tau-emotion-bench/Tracer-High-4B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tau-emotion-bench/Tracer-High-4B with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="tau-emotion-bench/Tracer-High-4B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("tau-emotion-bench/Tracer-High-4B")
model = AutoModelForCausalLM.from_pretrained("tau-emotion-bench/Tracer-High-4B", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use tau-emotion-bench/Tracer-High-4B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tau-emotion-bench/Tracer-High-4B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tau-emotion-bench/Tracer-High-4B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tau-emotion-bench/Tracer-High-4B

SGLang

How to use tau-emotion-bench/Tracer-High-4B with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "tau-emotion-bench/Tracer-High-4B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tau-emotion-bench/Tracer-High-4B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "tau-emotion-bench/Tracer-High-4B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tau-emotion-bench/Tracer-High-4B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tau-emotion-bench/Tracer-High-4B with Docker Model Runner:
```
docker model run hf.co/tau-emotion-bench/Tracer-High-4B
```

Tracer-High-4B — Interactive Tool-Using Agent

Tracer-High-4B is a multi-turn interactive tool-using agent fine-tuned from Qwen/Qwen3-4B using supervised fine-tuning on tool-use dialogues with emotionally expressive users.

Paper: τ-Emotion-Bench: Exposing the Interaction Fragility of LLM Agents through Real-World User Emotions
Project Page: τ-Emotion-Bench
Training Framework: TRL
Benchmark: τ-Emotion-Bench

Highlights

Trained on ten customer-service domains with emotionally expressive user behaviour.
Reasons in a <think> block before every reply or tool call; tool calls follow the Hermes convention.

Performance

Results on the ten domains of τ-Emotion-Bench under two user simulators. Scores are pass^k in %; the better of Qwen3-4B and Tracer-High-4B on each metric is in bold.

User Simulator: Qwen3-235B-A22B-Instruct-2507

Domain	Qwen3-4B pass^1	Tracer-High-4B pass^1	Qwen3-4B pass^3	Tracer-High-4B pass^3
Airline	18.6	42.9	2.9	22.9
Retail	20.0	56.7	1.4	30.0
Telecom	34.3	37.6	8.6	8.6
Telehealth	36.7	51.9	14.3	22.9
Bank	17.3	46.7	4.0	26.0
Restaurant	12.7	24.7	0.0	6.0
Movie	2.0	9.3	0.0	2.0
School	16.0	51.3	0.0	34.0
Travel	4.7	13.3	0.0	0.0
Basketball	17.3	42.7	8.0	20.0
Average	18.0	37.7	3.9	17.2

User Simulator: GLM-5.2

Domain	Qwen3-4B pass^1	Tracer-High-4B pass^1	Qwen3-4B pass^3	Tracer-High-4B pass^3
Airline	31.0	59.5	20.0	41.4
Retail	51.0	80.5	27.1	67.1
Telecom	65.7	74.3	48.6	52.9
Telehealth	62.9	78.6	44.3	62.9
Bank	22.0	46.0	6.0	26.0
Restaurant	23.3	35.3	6.0	20.0
Movie	2.7	16.0	0.0	6.0
School	17.3	57.3	2.0	40.0
Travel	11.3	21.3	2.0	2.0
Basketball	25.3	44.7	6.0	24.0
Average	31.2	51.4	16.2	34.2

Training

Method

Data: Multi-turn tool-use dialogues with emotionally expressive users across ten domains (Retail, Telehealth, Telecom, Airline, Bank, Basketball, Movie, Restaurant, School, Travel), released as Tracer-High-tau-emotion-data.
Supervised Fine-Tuning (SFT): The base model is fine-tuned on turn-level samples in which the target assistant turn contains an inline <think> block followed by the reply or tool call.

Infrastructure

Training was run on 2× NVIDIA H100 NVL with FSDP/DDP data parallelism, FlashAttention-3, and gradient checkpointing.

Hyperparameters

	SFT
Effective Batch Size	32
Learning Rate	5e-5
LR Schedule	cosine, 3% warmup
Optimizer	paged_adamw_8bit
Epochs / Steps	3 epochs / 3,042 steps
Max Context Length	32,768
Precision	bf16

Training Data

Dataset	Description	Samples
Tracer-High-tau-emotion-data	SFT training data (ten domains)	32,740

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "tau-emotion-bench/Tracer-High-4B"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, dtype="bfloat16", device_map="auto")

tools = [{
    "type": "function",
    "function": {
        "name": "get_reservation_details",
        "description": "Get the details of a reservation.",
        "parameters": {"type": "object",
                       "properties": {"reservation_id": {"type": "string"}},
                       "required": ["reservation_id"]},
    },
}]
messages = [
    {"role": "system", "content": "<domain policy>"},
    {"role": "user", "content": "I need to cancel my reservation M9OL3W."},
]
inputs = tok.apply_chat_template(messages, tools=tools, add_generation_prompt=True,
                                 return_tensors="pt", return_dict=True).to(model.device)
out = model.generate(**inputs, max_new_tokens=1024)
print(tok.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Serving with vLLM:

vllm serve tau-emotion-bench/Tracer-High-4B \
  --dtype bfloat16 --max-model-len 32768 \
  --enable-auto-tool-choice --tool-call-parser hermes --reasoning-parser qwen3

Keep thinking on and use the model's sampling defaults in the request body:

{"temperature": 0.6, "top_p": 0.95, "top_k": 20, "min_p": 0,
 "chat_template_kwargs": {"enable_thinking": true}}

Citation

@article{singh2026tauemotionbench,
  title={$\tau$-Emotion-Bench: Exposing the Interaction Fragility of LLM Agents through Real-World User Emotions},
  author={Singh, Rishitosh Kumar and Saeidi, Amir and Hegde, Shamanthak and Kumbhar, Shrinidhi and Baral, Chitta},
  journal={arXiv preprint arXiv:2610.19560},
  year={2026},
  url={https://arxiv.org/abs/2610.19560}
}