Instructions to use iaroy/sgc-522m-sft with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use iaroy/sgc-522m-sft with Transformers:

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

pipe = pipeline("text-generation", model="iaroy/sgc-522m-sft", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("iaroy/sgc-522m-sft", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use iaroy/sgc-522m-sft with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "iaroy/sgc-522m-sft"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "iaroy/sgc-522m-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/iaroy/sgc-522m-sft

SGLang

How to use iaroy/sgc-522m-sft 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 "iaroy/sgc-522m-sft" \
    --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": "iaroy/sgc-522m-sft",
		"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 "iaroy/sgc-522m-sft" \
        --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": "iaroy/sgc-522m-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use iaroy/sgc-522m-sft with Docker Model Runner:
```
docker model run hf.co/iaroy/sgc-522m-sft
```

sgc-522m-sft

sgc (Single-GPU Coder) is a 522M-parameter decoder-only transformer for Python and C++ code, pretrained from scratch on 103B tokens in 76.3 hours on one NVIDIA GH200 GPU. This repository holds the model after supervised fine-tuning (SFT) on reasoning traces, before reinforcement learning. The final model is sgc-522m.

Paper: One GPU, 0.1 Trillion Tokens: Pretraining a Sub-Billion-Parameter Code Model
Code, training and evaluation records: github.com/indrajeetadityaroy9/single-gpu-code-llm

Results

Model	HumanEval	HumanEval+
sgc-522m-base	26.8	22.0
sgc-522m-sft	32.9	31.7
sgc-522m	40.9	38.4

pass@1 with greedy decoding, scored by EvalPlus. On 512 held-out KodCode problems (4 samples each at temperature 0.6, top-p 0.95, up to 16,384 tokens), this model's pass@1 is 0.334, and 918 of its 2,048 responses do not finish: most fall into a repeating loop. RL reduces this to 266 (sgc-522m).

Prompt format and usage

As for sgc-522m: the chat template renders <|bos|>, the user message, then \n\n<think>\n; the model reasons, closes with </think> and answers. Serve it with the repository's scripts/serve.sh iaroy/sgc-522m-sft, or load it with transformers (trust_remote_code=True).

Training

SFT for 2 epochs (7,676 steps, 11.32B tokens, 11.2 hours) of reasoning traces from Nemotron-Cascade SFT (code), Nemotron SFT Competitive Programming v2 and KodCode-V1-SFT-R1, starting from sgc-522m-base.

Downloads last month: 403

Safetensors

Model size

0.5B params

Tensor type

F32

BF16

Model tree for iaroy/sgc-522m-sft

Base model

iaroy/sgc-522m-base

Finetuned

(1)

this model

Finetunes

1 model

iaroy
/

sgc-522m-sft

sgc-522m-sft

Results

Prompt format and usage

Training

Model tree for iaroy/sgc-522m-sft

Datasets used to train iaroy/sgc-522m-sft