Instructions to use datajuicer/LLaMA-1B-dj-refine-150B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use datajuicer/LLaMA-1B-dj-refine-150B with Transformers:

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

pipe = pipeline("text-generation", model="datajuicer/LLaMA-1B-dj-refine-150B")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("datajuicer/LLaMA-1B-dj-refine-150B")
model = AutoModelForCausalLM.from_pretrained("datajuicer/LLaMA-1B-dj-refine-150B", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use datajuicer/LLaMA-1B-dj-refine-150B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "datajuicer/LLaMA-1B-dj-refine-150B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "datajuicer/LLaMA-1B-dj-refine-150B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/datajuicer/LLaMA-1B-dj-refine-150B

SGLang

How to use datajuicer/LLaMA-1B-dj-refine-150B 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 "datajuicer/LLaMA-1B-dj-refine-150B" \
    --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": "datajuicer/LLaMA-1B-dj-refine-150B",
		"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 "datajuicer/LLaMA-1B-dj-refine-150B" \
        --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": "datajuicer/LLaMA-1B-dj-refine-150B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use datajuicer/LLaMA-1B-dj-refine-150B with Docker Model Runner:
```
docker model run hf.co/datajuicer/LLaMA-1B-dj-refine-150B
```

yxdyc commited on Oct 30, 2023

Commit

94ed815

1 Parent(s): 2ecfea2

Update README.md

Browse files

Files changed (1) hide show

README.md +27 -0

README.md CHANGED Viewed

@@ -1,3 +1,30 @@
 ---
 license: apache-2.0
 ---

 ---
 license: apache-2.0
+datasets:
+- datajuicer/redpajama-wiki-refined-by-data-juicer
+- datajuicer/redpajama-arxiv-refined-by-data-juicer
+- datajuicer/redpajama-c4-refined-by-data-juicer
+- datajuicer/redpajama-book-refined-by-data-juicer
+- datajuicer/redpajama-cc-2019-30-refined-by-data-juicer
+- datajuicer/redpajama-cc-2020-05-refined-by-data-juicer
+- datajuicer/redpajama-cc-2021-04-refined-by-data-juicer
+- datajuicer/redpajama-cc-2022-05-refined-by-data-juicer
+- datajuicer/redpajama-cc-2023-06-refined-by-data-juicer
+- datajuicer/redpajama-pile-stackexchange-refined-by-data-juicer
+- datajuicer/redpajama-stack-code-refined-by-data-juicer
+- datajuicer/the-pile-nih-refined-by-data-juicer
+- datajuicer/the-pile-europarl-refined-by-data-juicer
+- datajuicer/the-pile-philpaper-refined-by-data-juicer
+- datajuicer/the-pile-pubmed-abstracts-refined-by-data-juicer
+- datajuicer/the-pile-pubmed-central-refined-by-data-juicer
+- datajuicer/the-pile-freelaw-refined-by-data-juicer
+- datajuicer/the-pile-hackernews-refined-by-data-juicer
 ---
+This is a reference LLM from [Data-Juicer](https://github.com/alibaba/data-juicer).
+The model architecture is LLaMA-1.3B and we adopt the [OpenLLaMA](https://github.com/openlm-research/open_llama) implementation.
+The model is pre-trained on 150B tokens of Data-Juicer's refined RedPajama and Pile.
+It achieves an average score of 34.21 over 16 HELM tasks, beating Falcon-1.3B (trained on 350B tokens from RefinedWeb), Pythia-1.4B (trained on 300B tokens from original Pile) and Open-LLaMA-1.3B (trained on 150B tokens from original RedPajama and Pile).
+For more details, please refer to our [paper](https://arxiv.org/abs/2309.02033).