Instructions to use OKoshi/stage2-02 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use OKoshi/stage2-02 with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("/content/sft_work/stage1_merged")
model = PeftModel.from_pretrained(base_model, "OKoshi/stage2-02")

Transformers

How to use OKoshi/stage2-02 with Transformers:

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

pipe = pipeline("text-generation", model="OKoshi/stage2-02")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("OKoshi/stage2-02", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use OKoshi/stage2-02 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "OKoshi/stage2-02"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "OKoshi/stage2-02",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/OKoshi/stage2-02

SGLang

How to use OKoshi/stage2-02 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 "OKoshi/stage2-02" \
    --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": "OKoshi/stage2-02",
		"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 "OKoshi/stage2-02" \
        --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": "OKoshi/stage2-02",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use OKoshi/stage2-02 with Docker Model Runner:
```
docker model run hf.co/OKoshi/stage2-02
```

GENIAC-PRIZE 2026 — Stage 2 Adapter

Model Description

This repository contains the Stage 2 LoRA adapter developed for GENIAC-PRIZE 2026 Theme 2 (Mathematical LLM).

Base model: Qwen3-4B-Base
Training method: Supervised Fine-Tuning (SFT) with LoRA
Training dataset: LIMO
Dataset: GAIR/LIMO

Training Data and License

The LIMO dataset was used for Stage 2 training. Refer to the dataset repository for its license and usage terms.

Intended Use

This adapter is intended for use in the GENIAC-PRIZE 2026 Theme 2 mathematical language model development task.

Usage

Load this adapter using the appropriate base model and the Stage 1 checkpoint or merged model configuration required by the training pipeline.

Notes

This repository contains the Stage 2 adapter weights, not a standalone merged model. The Stage 2 adapter was trained after Stage 1.

Downloads last month: 24

Model tree for OKoshi/stage2-02

Base model

Qwen/Qwen3-4B-Base

Quantized

unsloth/Qwen3-4B-Base-unsloth-bnb-4bit

Adapter

(23)

this model