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

Libraries

How to use TheBloke/phi-2-GGUF with Transformers:

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

pipe = pipeline("text-generation", model="TheBloke/phi-2-GGUF")

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("TheBloke/phi-2-GGUF", dtype="auto")

llama-cpp-python

How to use TheBloke/phi-2-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="TheBloke/phi-2-GGUF",
	filename="phi-2.Q2_K.gguf",
)

output = llm(
	"Once upon a time,",
	max_tokens=512,
	echo=True
)
print(output)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use TheBloke/phi-2-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf TheBloke/phi-2-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf TheBloke/phi-2-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf TheBloke/phi-2-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf TheBloke/phi-2-GGUF:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf TheBloke/phi-2-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf TheBloke/phi-2-GGUF:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf TheBloke/phi-2-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf TheBloke/phi-2-GGUF:Q4_K_M

Use Docker

docker model run hf.co/TheBloke/phi-2-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use TheBloke/phi-2-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "TheBloke/phi-2-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "TheBloke/phi-2-GGUF",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/TheBloke/phi-2-GGUF:Q4_K_M

SGLang

How to use TheBloke/phi-2-GGUF 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 "TheBloke/phi-2-GGUF" \
    --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": "TheBloke/phi-2-GGUF",
		"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 "TheBloke/phi-2-GGUF" \
        --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": "TheBloke/phi-2-GGUF",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Ollama
How to use TheBloke/phi-2-GGUF with Ollama:
```
ollama run hf.co/TheBloke/phi-2-GGUF:Q4_K_M
```

Unsloth Studio new

How to use TheBloke/phi-2-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for TheBloke/phi-2-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for TheBloke/phi-2-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for TheBloke/phi-2-GGUF to start chatting

Docker Model Runner
How to use TheBloke/phi-2-GGUF with Docker Model Runner:
```
docker model run hf.co/TheBloke/phi-2-GGUF:Q4_K_M
```

Lemonade

How to use TheBloke/phi-2-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull TheBloke/phi-2-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.phi-2-GGUF-Q4_K_M

List all available models

lemonade list

today's version of llama.cpp results in an error

by LaferriereJC - opened Dec 19, 2023

Discussion

LaferriereJC

Dec 19, 2023

error loading model: unknown model architecture: 'phi2'
llama_load_model_from_file: failed to load model
AVX = 1 | AVX2 = 1 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 |
2023-12-18 16:57:06 ERROR:Failed to load the model.
Traceback (most recent call last):
File "/data/text-generation-webui/modules/ui_model_menu.py", line 209, in load_model_wrapper
shared.model, shared.tokenizer = load_model(selected_model, loader)
File "/data/text-generation-webui/modules/models.py", line 89, in load_model
output = load_func_maploader
File "/data/text-generation-webui/modules/models.py", line 259, in llamacpp_loader
model, tokenizer = LlamaCppModel.from_pretrained(model_file)
File "/data/text-generation-webui/modules/llamacpp_model.py", line 91, in from_pretrained
result.model = Llama(**params)
File "/data/llama-cpp-python/llama_cpp/llama.py", line 963, in init
self._n_vocab = self.n_vocab()
File "/data/llama-cpp-python/llama_cpp/llama.py", line 2270, in n_vocab
return self._model.n_vocab()
File "/data/llama-cpp-python/llama_cpp/llama.py", line 252, in n_vocab
assert self.model is not None
AssertionError

ociubotaru

Dec 19, 2023

•

edited Dec 19, 2023

I had the same issue using LM studio 0.29
{
"cause": {
"cause": "unknown model architecture: 'phi2'",
"title": "Failed to load model",
"errorData": {
"n_ctx": 2048,
"n_batch": 512,
"n_gpu_layers": 1
}
},

Downloaded and installed the Beta V8 version (https://lmstudio.ai/beta-releases.html) and the problem was solved.

Pumba2

Dec 19, 2023

•

edited Dec 20, 2023

I knew there would be a problem. Just had to :D Every time there is a good model it either doesnt work properly or just doesnt work. At least we got the quantized ver so in the future the llm model apps will fix it.

dyoung

Jan 4, 2024

I think the newest version of llama-cpp works for phi-2 now. The community was hard at work with making adjustments for it's slight differences during the month. (https://github.com/ggerganov/llama.cpp/issues/4437) Just likely need to pip update or remake your build for the latest.

dyoung

Jan 4, 2024

•

edited Jan 4, 2024

Can't speak for LM studio. I'm not a user of it. Sorry. Though it's good to see that something can load phi-2 gguf quants such as LM Studio.

mox

Jan 7, 2024

•

edited Jan 7, 2024

Can confirm that it worked with upgrading on MAC: CMAKE_ARGS="-DLLAMA_METAL=on" pip install -U llama-cpp-python

dyoung

Jan 16, 2024

@mox Same with my side. I've successfully used phi with the llama-cpp family shorty after my 1st reply 12 days ago. As well as several times since. Feels like a old hat now. Been pretty happy with it.
@LaferriereJC I've not checked, but has LM Studio got it working yet for phi? Likely has. It not, that's a bummer. And really sorry. The wait can be painful sometimes.

NikeshK

Apr 4, 2024

Has anyone made it work with CTransformers as i am getting a error for model_type = 'phi-msft'

YaTharThShaRma999

Apr 4, 2024

@NikeshK ctransformers is pretty outdated now. Use something like llama cpp python which is maintained if you want to use llama cpp with python

dyoung

Apr 5, 2024

Has anyone made it work with CTransformers as i am getting a error for model_type = 'phi-msft'

ollama is a good alternative as well. Especially if you wanted to keep api use simple like ctransformers was trying to do. ollama has llama.cpp as the core. And it works on all common consumer os's (linux, mac and windows). As well as common consumer computing acceleration hardware (nvidia, amd, apple MPS, ect.).

Here is the links:

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