Instructions to use schwp/TinyMathLlama-1.1B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use schwp/TinyMathLlama-1.1B with Transformers:

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

pipe = pipeline("text-generation", model="schwp/TinyMathLlama-1.1B")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("schwp/TinyMathLlama-1.1B")
model = AutoModelForCausalLM.from_pretrained("schwp/TinyMathLlama-1.1B", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use schwp/TinyMathLlama-1.1B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "schwp/TinyMathLlama-1.1B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "schwp/TinyMathLlama-1.1B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/schwp/TinyMathLlama-1.1B

SGLang

How to use schwp/TinyMathLlama-1.1B 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 "schwp/TinyMathLlama-1.1B" \
    --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": "schwp/TinyMathLlama-1.1B",
		"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 "schwp/TinyMathLlama-1.1B" \
        --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": "schwp/TinyMathLlama-1.1B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use schwp/TinyMathLlama-1.1B with Docker Model Runner:
```
docker model run hf.co/schwp/TinyMathLlama-1.1B
```

TinyMathLlama-1.1B

File size: 914 Bytes

e149c9f
 
c497546
 
 
 
 
 
 
 
 
 
e149c9f
 
c497546
e149c9f
c497546
e149c9f
c497546
 
 
 
e149c9f
c497546
e149c9f
c497546
 
e149c9f
c497546
e149c9f
c497546

---
library_name: transformers
base_model: TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T
datasets:
- meta-math/MetaMathQA
- openai/gsm8k
tags:
- lora
- math
- fine-tuned
language:
- en
---

# TinyMathLlama-1.1B

TinyLlama-1.1B fine-tuned on MetaMathQA using a from-scratch LoRA implementation.

- **LoRA config:** r=8, $\alpha$=16, target modules: q_proj + v_proj
- **Training data:** 10k samples from MetaMathQA
- **GSM8K accuracy:** 3.0% (base: 1.5%, 2x improvement)
- **Trainable params:** 1,126,400 / 1,101,174,784 (0.1%)

## Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("schwp/schwp/TinyMathLlama-1.1B")

## Training & Evaluation

The training and evaluation scripts are available on this [github repository](https://github.com/schwp/lora-from-scratch).
The whole LoRA implementation used for the fine-tuning is also on the repository.