Instructions to use dphn/dolphin-phi-2-kensho with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use dphn/dolphin-phi-2-kensho with Transformers:

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

pipe = pipeline("text-generation", model="dphn/dolphin-phi-2-kensho", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("dphn/dolphin-phi-2-kensho", trust_remote_code=True, dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use dphn/dolphin-phi-2-kensho with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "dphn/dolphin-phi-2-kensho"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "dphn/dolphin-phi-2-kensho",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/dphn/dolphin-phi-2-kensho

SGLang

How to use dphn/dolphin-phi-2-kensho 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 "dphn/dolphin-phi-2-kensho" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "dphn/dolphin-phi-2-kensho",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "dphn/dolphin-phi-2-kensho" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "dphn/dolphin-phi-2-kensho",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use dphn/dolphin-phi-2-kensho with Docker Model Runner:
```
docker model run hf.co/dphn/dolphin-phi-2-kensho
```

ZennyKenny commited on Mar 5, 2024

Commit

10ccf92

verified ·

1 Parent(s): ea7a4cb

Update arxiv.org link

Browse files

As it was, [https://arxiv.org/abs/2402.14905] (no whitespace trailing URL) appends URL-encoded square bracket (%5D) to arxiv URL and breaks it. Just added whitespace so that link works.

Files changed (1) hide show

README.md +1 -1

README.md CHANGED Viewed

@@ -5,7 +5,7 @@
 By Fernando, Eric and David
-This is a hack around pytorch + huggingface Transformers library to make the original Dolphin Phi-2 to behave in a way inspired by the Meta's paper "MobileLLM: Optimizing Sub-billion Parameter Language Models for On-Device Use Cases" [https://arxiv.org/abs/2402.14905]
 One of the key ideas is that it works as if it was like "an online passthrough", by applying a loop on a module SuperClass, that groups layers, in a such way they get their forward method repeated in a loop.
 So, in theory, you can observe more intelligence in the same way MegaDolphin 120b, Professor 155b, Venus120b and other huge models, but use way less vRAM, because instead of cloning the weights, we share them in the vRAM.

 By Fernando, Eric and David
+This is a hack around pytorch + huggingface Transformers library to make the original Dolphin Phi-2 to behave in a way inspired by the Meta's paper "MobileLLM: Optimizing Sub-billion Parameter Language Models for On-Device Use Cases" [ https://arxiv.org/abs/2402.14905 ]
 One of the key ideas is that it works as if it was like "an online passthrough", by applying a loop on a module SuperClass, that groups layers, in a such way they get their forward method repeated in a loop.
 So, in theory, you can observe more intelligence in the same way MegaDolphin 120b, Professor 155b, Venus120b and other huge models, but use way less vRAM, because instead of cloning the weights, we share them in the vRAM.