Instructions to use Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo with Transformers:

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

pipe = pipeline("text-generation", model="Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo", trust_remote_code=True)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo

SGLang

How to use Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo 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 "Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo" \
    --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": "Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo",
		"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 "Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo" \
        --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": "Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo with Docker Model Runner:
```
docker model run hf.co/Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo
```

Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo

This repo contains a full merged model produced from Dapinsky/PIT-4B-FT-202212-math-reasoning-sft and a PEFT LoRA adapter. It is intended to be loadable with the same Transformers API used for the base Diamegs PIT model:

from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo", trust_remote_code=True)

Training Summary

Base model: Dapinsky/PIT-4B-FT-202212-math-reasoning-sft
Fine-tuning method: LoRA post-training, merged into the base weights for upload.
Tokenizer files copied from: base model

LoRA Configuration

{
  "peft_type": "LORA",
  "task_type": "CAUSAL_LM",
  "r": 16,
  "lora_alpha": 32,
  "lora_dropout": 0.05,
  "target_modules": [
    "c_fc",
    "c_proj",
    "c_v",
    "c_k",
    "c_q"
  ],
  "base_model_name_or_path": "Dapinsky/PIT-4B-FT-202212-math-reasoning-sft"
}

Downloads last month: 12

Safetensors

Model size

4B params

Tensor type

F32

Model tree for Dapinsky/PIT-4B-FT-202212-math-reasoning-dpo

Base model

Diamegs/PIT-4B-202212

Finetuned

Diamegs/PIT-4B-FT-202212

Finetuned

Dapinsky/PIT-4B-FT-202212-math-reasoning-sft

Finetuned

(1)

this model