Instructions to use BitZone910/Trace-OPSD-Qwen2.5-3B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use BitZone910/Trace-OPSD-Qwen2.5-3B with Transformers:

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

pipe = pipeline("text-generation", model="BitZone910/Trace-OPSD-Qwen2.5-3B")
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("BitZone910/Trace-OPSD-Qwen2.5-3B")
model = AutoModelForCausalLM.from_pretrained("BitZone910/Trace-OPSD-Qwen2.5-3B", 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 BitZone910/Trace-OPSD-Qwen2.5-3B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "BitZone910/Trace-OPSD-Qwen2.5-3B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "BitZone910/Trace-OPSD-Qwen2.5-3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/BitZone910/Trace-OPSD-Qwen2.5-3B

SGLang

How to use BitZone910/Trace-OPSD-Qwen2.5-3B 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 "BitZone910/Trace-OPSD-Qwen2.5-3B" \
    --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": "BitZone910/Trace-OPSD-Qwen2.5-3B",
		"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 "BitZone910/Trace-OPSD-Qwen2.5-3B" \
        --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": "BitZone910/Trace-OPSD-Qwen2.5-3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use BitZone910/Trace-OPSD-Qwen2.5-3B with Docker Model Runner:
```
docker model run hf.co/BitZone910/Trace-OPSD-Qwen2.5-3B
```

Trace-OPSD-Qwen2.5-3B

Built with Qwen. A ToolElite mathematical tool-use model, initialized from SFT-Qwen2.5-3B and trained with TRACE-OPSD.

Training provenance

Original model: Qwen2.5-3B-Instruct.
RL data: rl_trace_opsd,

Loading

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "BitZone910/Trace-OPSD-Qwen2.5-3B"
tokenizer = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
    repo, torch_dtype=torch.bfloat16, device_map="auto"
)

Tool use

The model was trained with Python execution and General Search. Tool calls require an external executor and retrieval service; model.generate() does not execute tools. ToolElite uses XML actions such as <python>...</python> and <search>...</search>, tool feedback, and \\boxed{...} final answers. Use the ToolElite agent loop and its tool prompt to reproduce the evaluation protocol.

Attribution and license

Qwen weights and derivatives are subject to the included Qwen Research License, which restricts use to non-commercial research/evaluation unless separately licensed. See LICENSE and NOTICE; this release does not relicense the model under MIT or Apache-2.0.

The RL source is derived from NuminaMath-TIR, with Trace-OPSD difficulty labels, solution hindsight, and fixed train/validation splits.

Downloads last month: 225

Safetensors

Model size

3B params

Tensor type

BF16

Model tree for BitZone910/Trace-OPSD-Qwen2.5-3B

Base model

Qwen/Qwen2.5-3B

Finetuned

Qwen/Qwen2.5-3B-Instruct

Finetuned

BitZone910/SFT-Qwen2.5-3B

Finetuned

(1)

this model

BitZone910
/

Trace-OPSD-Qwen2.5-3B

Trace-OPSD-Qwen2.5-3B

Training provenance

Loading

Tool use

Attribution and license

Model tree for BitZone910/Trace-OPSD-Qwen2.5-3B

Dataset used to train BitZone910/Trace-OPSD-Qwen2.5-3B