Instructions to use RedHatAI/Phi-3-mini-128k-instruct-FP8 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use RedHatAI/Phi-3-mini-128k-instruct-FP8 with Transformers:

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

pipe = pipeline("text-generation", model="RedHatAI/Phi-3-mini-128k-instruct-FP8", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("RedHatAI/Phi-3-mini-128k-instruct-FP8", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("RedHatAI/Phi-3-mini-128k-instruct-FP8", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use RedHatAI/Phi-3-mini-128k-instruct-FP8 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "RedHatAI/Phi-3-mini-128k-instruct-FP8"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "RedHatAI/Phi-3-mini-128k-instruct-FP8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/RedHatAI/Phi-3-mini-128k-instruct-FP8

SGLang

How to use RedHatAI/Phi-3-mini-128k-instruct-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 "RedHatAI/Phi-3-mini-128k-instruct-FP8" \
    --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": "RedHatAI/Phi-3-mini-128k-instruct-FP8",
		"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 "RedHatAI/Phi-3-mini-128k-instruct-FP8" \
        --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": "RedHatAI/Phi-3-mini-128k-instruct-FP8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use RedHatAI/Phi-3-mini-128k-instruct-FP8 with Docker Model Runner:
```
docker model run hf.co/RedHatAI/Phi-3-mini-128k-instruct-FP8
```

Prompt format in `tokenizer_config.json`

by justinthelaw - opened Aug 9, 2024

Discussion

justinthelaw

Aug 9, 2024

•

edited Aug 9, 2024

The prompt format in the tokenizer_config.json seems to vary with the other Phi-3 instruct variants in the wild and from the official source (Microsoft). The change seems to be the removal of the system prompt section and conditional logic.

NeuralMagic: https://huggingface.co/neuralmagic/Phi-3-mini-128k-instruct-FP8/blob/402f133d5636dcb6e0cbb1209663b8956b36d0be/tokenizer_config.json#L119

Upstream: https://huggingface.co/microsoft/Phi-3-mini-128k-instruct/blob/d548c233192db00165d842bf8edff054bb3212f8/tokenizer_config.json#L119

Is this an intentional change? If so, why? I believe the system prompt is generally useful for most cases, like when creating agents or assistants.

markurtz

Red Hat AI org Aug 9, 2024

Thanks @justinthelaw , we are taking a look at these today to rectify!

justinthelaw

Aug 9, 2024

•

edited Aug 9, 2024

Additionally, I noticed the actual config definition code is old and out-of-date from upstream, leading to the old missing transformers_modules issue seen by the older version of Phi-3 when used with vLLM (0.4.x - 0.5.x). Recommend adding the configuration_phi3.py and modeling_phi3.py from upstream as an import in the config.json or direct download into this repo.

ModuleNotFoundError: No module named 'transformers_modules'

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/home/nonroot/.pyenv/versions/3.11.6/lib/python3.11/asyncio/events.py", line 80, in _run
    self._context.run(self._callback, *self._args)
  File "/home/leapfrogai/.venv/lib/python3.11/site-packages/vllm/engine/async_llm_engine.py", line 59, in _log_task_completion
    raise AsyncEngineDeadError(
vllm.engine.async_llm_engine.AsyncEngineDeadError: Task finished unexpectedly. This should never happen! Please open an issue on Github. See stack trace above for theactual cause.

Lin-K76

Aug 11, 2024

Hi @justinthelaw , thanks! The new config files should handle this correctly now.

Lin-K76 changed discussion status to closed Aug 11, 2024

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