Instructions to use moekotb12345/SmolLM3-3B-GSM8K-QLoRA with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use moekotb12345/SmolLM3-3B-GSM8K-QLoRA with Transformers:

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

pipe = pipeline("text-generation", model="moekotb12345/SmolLM3-3B-GSM8K-QLoRA")

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

tokenizer = AutoTokenizer.from_pretrained("moekotb12345/SmolLM3-3B-GSM8K-QLoRA")
model = AutoModelForCausalLM.from_pretrained("moekotb12345/SmolLM3-3B-GSM8K-QLoRA", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use moekotb12345/SmolLM3-3B-GSM8K-QLoRA with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "moekotb12345/SmolLM3-3B-GSM8K-QLoRA"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "moekotb12345/SmolLM3-3B-GSM8K-QLoRA",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/moekotb12345/SmolLM3-3B-GSM8K-QLoRA

SGLang

How to use moekotb12345/SmolLM3-3B-GSM8K-QLoRA 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 "moekotb12345/SmolLM3-3B-GSM8K-QLoRA" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "moekotb12345/SmolLM3-3B-GSM8K-QLoRA",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "moekotb12345/SmolLM3-3B-GSM8K-QLoRA" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "moekotb12345/SmolLM3-3B-GSM8K-QLoRA",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use moekotb12345/SmolLM3-3B-GSM8K-QLoRA with Docker Model Runner:
```
docker model run hf.co/moekotb12345/SmolLM3-3B-GSM8K-QLoRA
```

SmolLM3-3B-GSM8K-QLoRA

Unit 1 project for the Hugging Face smol course. HuggingFaceTB/SmolLM3-3B-Base fine-tuned with QLoRA (TRL SFT) on step-by-step GSM8K solutions, formatted exactly like the lighteval|gsm8k|0|0 prompt (Question: ...\nAnswer:), ending with The final answer is N. so answers fit in the eval's 256-token budget. LoRA adapter merged into bf16 weights.

Training

Data: GSM8K train split only (calculator annotations removed); 300 examples held out for validation
QLoRA: 4-bit NF4 base, LoRA r=16, alpha=32, dropout 0.05, all linear layers
2 epochs, lr 0.0002, cosine schedule, effective batch 16, max length 512, fp16 on a free Colab T4
Loss on completions only; final validation loss 0.530

Quick check (first 200 GSM8K test questions, greedy, 256 tokens, 4-bit)

Model	Accuracy
HuggingFaceTB/SmolLM3-3B-Base	3.5%
This model	70.0%

Official leaderboard number: see the lighteval results dataset moekotb12345/details_moekotb12345__SmolLM3-3B-GSM8K-QLoRA_private.

Usage

from transformers import pipeline
pipe = pipeline("text-generation", model="moekotb12345/SmolLM3-3B-GSM8K-QLoRA")
print(pipe("Question: A pen costs $3 and a notebook costs twice as much. How much for 2 of each?\nAnswer:",
           max_new_tokens=256, stop_strings=["Question:"], tokenizer=pipe.tokenizer)[0]["generated_text"])

Downloads last month: 562

Safetensors

Model size

3B params

Tensor type

BF16

Model tree for moekotb12345/SmolLM3-3B-GSM8K-QLoRA

Base model

HuggingFaceTB/SmolLM3-3B-Base

Finetuned

(106)

this model

moekotb12345
/

SmolLM3-3B-GSM8K-QLoRA

SmolLM3-3B-GSM8K-QLoRA

Training

Quick check (first 200 GSM8K test questions, greedy, 256 tokens, 4-bit)

Usage

Model tree for moekotb12345/SmolLM3-3B-GSM8K-QLoRA

Dataset used to train moekotb12345/SmolLM3-3B-GSM8K-QLoRA