Instructions to use joaoalvarenga/bloom-8bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use joaoalvarenga/bloom-8bit with Transformers:

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

pipe = pipeline("text-generation", model="joaoalvarenga/bloom-8bit")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("joaoalvarenga/bloom-8bit")
model = AutoModelForCausalLM.from_pretrained("joaoalvarenga/bloom-8bit", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use joaoalvarenga/bloom-8bit with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/joaoalvarenga/bloom-8bit

SGLang

How to use joaoalvarenga/bloom-8bit 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 "joaoalvarenga/bloom-8bit" \
    --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": "joaoalvarenga/bloom-8bit",
		"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 "joaoalvarenga/bloom-8bit" \
        --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": "joaoalvarenga/bloom-8bit",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use joaoalvarenga/bloom-8bit with Docker Model Runner:
```
docker model run hf.co/joaoalvarenga/bloom-8bit
```

joaoalvarenga commited on Jul 12, 2022

Commit

8b76f74

1 Parent(s): 261fbda

Update README.md

Browse files

Files changed (1) hide show

README.md +2 -2

README.md CHANGED Viewed

@@ -50,13 +50,13 @@ pipeline_tag: text-generation
 ---
 ### Quantized bigscience/bloom with 8-bit weights
-Heavily inspired by [Hivemind's GPT-J-6B with 8-bit weights](https://huggingface.co/hivemind/gpt-j-6B-8bit), this is a version of [bigscience/bloom](https://huggingface.co/bigscience/bloom) a ~176 billions parameters language model that you run and fine-tune with less memory.
 Here, we also apply [LoRA (Low Rank Adapters)](https://arxiv.org/abs/2106.09685) to reduce model size. The original version takes \~353GB memory, this version takes **\~180GB**.
 Our main goal is to generate a model compressed enough to be deployed in a traditional Kubernetes cluster.
-### How to fine tune
 In this [notebook](https://nbviewer.org/urls/huggingface.co/joaoalvarenga/bloom-8bit/raw/main/fine-tuning-example.ipynb) you can find an adaptation from [Hivemind's GPT-J 8-bit fine-tuning notebook](https://colab.research.google.com/drive/1ft6wQU0BhqG5PRlwgaZJv2VukKKjU4Es) to fine-tune Bloom 8-bit with a 3x NVIDIA A100 instance.

 ---
 ### Quantized bigscience/bloom with 8-bit weights
+Heavily inspired by [Hivemind's GPT-J-6B with 8-bit weights](https://huggingface.co/hivemind/gpt-j-6B-8bit), this is a version of [bigscience/bloom](https://huggingface.co/bigscience/bloom) a ~176 billion parameters language model that you run and fine-tune with less memory.
 Here, we also apply [LoRA (Low Rank Adapters)](https://arxiv.org/abs/2106.09685) to reduce model size. The original version takes \~353GB memory, this version takes **\~180GB**.
 Our main goal is to generate a model compressed enough to be deployed in a traditional Kubernetes cluster.
+### How to fine-tune
 In this [notebook](https://nbviewer.org/urls/huggingface.co/joaoalvarenga/bloom-8bit/raw/main/fine-tuning-example.ipynb) you can find an adaptation from [Hivemind's GPT-J 8-bit fine-tuning notebook](https://colab.research.google.com/drive/1ft6wQU0BhqG5PRlwgaZJv2VukKKjU4Es) to fine-tune Bloom 8-bit with a 3x NVIDIA A100 instance.