Instructions to use stanfordnlp/backpack-gpt2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use stanfordnlp/backpack-gpt2 with Transformers:

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

pipe = pipeline("text-generation", model="stanfordnlp/backpack-gpt2", trust_remote_code=True)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("stanfordnlp/backpack-gpt2", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("stanfordnlp/backpack-gpt2", trust_remote_code=True)

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use stanfordnlp/backpack-gpt2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "stanfordnlp/backpack-gpt2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "stanfordnlp/backpack-gpt2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/stanfordnlp/backpack-gpt2

SGLang

How to use stanfordnlp/backpack-gpt2 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 "stanfordnlp/backpack-gpt2" \
    --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": "stanfordnlp/backpack-gpt2",
		"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 "stanfordnlp/backpack-gpt2" \
        --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": "stanfordnlp/backpack-gpt2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use stanfordnlp/backpack-gpt2 with Docker Model Runner:
```
docker model run hf.co/stanfordnlp/backpack-gpt2
```

Set more precise shape to the attention weights and outputs

by ivanzhouyq - opened May 29, 2023

base: refs/heads/main

←

from: refs/pr/1

Discussion Files changed

-2

ivanzhouyq

May 29, 2023

•

edited May 29, 2023

This PR sets more precise shape to the attention's weights, biases, and the outputs.

The original implementation assumes that the embedding size is a multiple of senses. However, when the hyperparamters are picked so that the embedding size is no longer the multiple of the number of senses, it would cause different number of parameters in encoded before and after reshaping. I found this mismatching when testing with a larger model with embedding size = 1280 and number of senses at 48. I received the following error:

- sense_weight_net.c_attn.bias: found shape torch.Size([2496]) in the checkpoint and torch.Size([2560]) in the model instantiated
- sense_weight_net.c_attn.weight: found shape torch.Size([2496, 1280]) in the checkpoint and torch.Size([2560, 1280]) in the model instantiated

This PR addresses this issue. Of course, a better solution should be recommending/enforcing the embedding size to be a full multiple of senses.

Set more precise shape to the attention weights and outputsb2c5167f

johnhew changed pull request status to merged Aug 14, 2023

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment