Instructions to use mgoin/Minitron-8B-Base-FP8 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mgoin/Minitron-8B-Base-FP8 with Transformers:

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

pipe = pipeline("text-generation", model="mgoin/Minitron-8B-Base-FP8")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("mgoin/Minitron-8B-Base-FP8")
model = AutoModelForCausalLM.from_pretrained("mgoin/Minitron-8B-Base-FP8", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use mgoin/Minitron-8B-Base-FP8 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "mgoin/Minitron-8B-Base-FP8"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mgoin/Minitron-8B-Base-FP8",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/mgoin/Minitron-8B-Base-FP8

SGLang

How to use mgoin/Minitron-8B-Base-FP8 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 "mgoin/Minitron-8B-Base-FP8" \
    --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": "mgoin/Minitron-8B-Base-FP8",
		"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 "mgoin/Minitron-8B-Base-FP8" \
        --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": "mgoin/Minitron-8B-Base-FP8",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use mgoin/Minitron-8B-Base-FP8 with Docker Model Runner:
```
docker model run hf.co/mgoin/Minitron-8B-Base-FP8
```

Minitron-8B-Base-FP8

FP8 quantized checkpoint of nvidia/Minitron-8B-Base for use with vLLM.

Evaluations

This quantized model:

lm_eval --model vllm --model_args pretrained=Minitron-8B-Base-FP8 --tasks gsm8k --num_fewshot 5 --batch_size auto

vllm (pretrained=Minitron-8B-Base-FP8), gen_kwargs: (None), limit: None, num_fewshot: 5, batch_size: auto
|Tasks|Version|     Filter     |n-shot|  Metric   |   |Value |   |Stderr|
|-----|------:|----------------|-----:|-----------|---|-----:|---|-----:|
|gsm8k|      3|flexible-extract|     5|exact_match|↑  |0.5019|±  |0.0138|
|     |       |strict-match    |     5|exact_match|↑  |0.4989|±  |0.0138|

Baseline:

lm_eval --model vllm --model_args pretrained=nvidia/Minitron-8B-Base --tasks gsm8k --num_fewshot 5 --batch_size auto

vllm (pretrained=nvidia/Minitron-8B-Base), gen_kwargs: (None), limit: None, num_fewshot: 5, batch_size: auto
|Tasks|Version|     Filter     |n-shot|  Metric   |   |Value |   |Stderr|
|-----|------:|----------------|-----:|-----------|---|-----:|---|-----:|
|gsm8k|      3|flexible-extract|     5|exact_match|↑  |0.5080|±  |0.0138|
|     |       |strict-match    |     5|exact_match|↑  |0.5064|±  |0.0138|

The original paper evals: