Instructions to use LoftQ/Phi-3-mini-4k-instruct-4bit-64rank with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use LoftQ/Phi-3-mini-4k-instruct-4bit-64rank with Transformers:

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

pipe = pipeline("text-generation", model="LoftQ/Phi-3-mini-4k-instruct-4bit-64rank", 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("LoftQ/Phi-3-mini-4k-instruct-4bit-64rank", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("LoftQ/Phi-3-mini-4k-instruct-4bit-64rank", 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 LoftQ/Phi-3-mini-4k-instruct-4bit-64rank with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "LoftQ/Phi-3-mini-4k-instruct-4bit-64rank"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "LoftQ/Phi-3-mini-4k-instruct-4bit-64rank",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/LoftQ/Phi-3-mini-4k-instruct-4bit-64rank

SGLang

How to use LoftQ/Phi-3-mini-4k-instruct-4bit-64rank 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 "LoftQ/Phi-3-mini-4k-instruct-4bit-64rank" \
    --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": "LoftQ/Phi-3-mini-4k-instruct-4bit-64rank",
		"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 "LoftQ/Phi-3-mini-4k-instruct-4bit-64rank" \
        --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": "LoftQ/Phi-3-mini-4k-instruct-4bit-64rank",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use LoftQ/Phi-3-mini-4k-instruct-4bit-64rank with Docker Model Runner:
```
docker model run hf.co/LoftQ/Phi-3-mini-4k-instruct-4bit-64rank
```

LoftQ Initialization

| Paper | Code | PEFT Example |

LoftQ (LoRA-fine-tuning-aware Quantization) provides a quantized backbone Q and LoRA adapters A and B, given a full-precision pre-trained weight W.

This model, LoftQ/Phi-3-mini-4k-instruct-4bit-64rank, is obtained from Phi-3-mini-4k-instruct. The backbone is under LoftQ/Phi-3-mini-4k-instruct-4bit-64rank and LoRA adapters are under the subfolder='loftq_init'.

Model Info

Backbone

Stored format: nf4
Size: ~ 2.5 GiB
Loaded format: bitsandbytes nf4
Size loaded on GPU: ~2.5 GiB

LoRA adapters

rank: 64
lora_alpha: 16
target_modules: ["qkv_proj", "o_proj", "up_gate_proj", "down_proj"]
rank_pattern: {"qkv_proj": 192, "up_gate_proj": 128}

Usage

Training Here's an example of loading this model and preparing for the LoRA fine-tuning.

import torch
from transformers import AutoModelForCausalLM, BitsAndBytesConfig
from peft import PeftModel

MODEL_ID = "LoftQ/Phi-3-mini-4k-instruct-4bit-64rank"

base_model = AutoModelForCausalLM.from_pretrained(MODEL_ID, trust_remote_code=True)
peft_model = PeftModel.from_pretrained(
    base_model,
    MODEL_ID,
    subfolder="loftq_init",
    is_trainable=True,
)

# Do training with peft_model ...

See the full code at our Github Repo

Citation

@article{li2023loftq,
  title={Loftq: Lora-fine-tuning-aware quantization for large language models},
  author={Li, Yixiao and Yu, Yifan and Liang, Chen and He, Pengcheng and Karampatziakis, Nikos and Chen, Weizhu and Zhao, Tuo},
  journal={arXiv preprint arXiv:2310.08659},
  year={2023}
}

Downloads last month: 2

Safetensors

Model size

4B params

Tensor type

F32

BF16

Collection including LoftQ/Phi-3-mini-4k-instruct-4bit-64rank

LoftQ Phi

Collection

LoftQ initialization of Phi-2 and Phi-3 • 3 items • Updated May 3, 2024 • 1

Paper for LoftQ/Phi-3-mini-4k-instruct-4bit-64rank

LoftQ: LoRA-Fine-Tuning-Aware Quantization for Large Language Models

Paper • 2310.08659 • Published Oct 12, 2023 • 29