microsoft/Phi-3-medium-128k-instruct · Edited starter script

Had a bit of trouble getting started, so here's my modified version with pip instructions. I ran this on an Ubuntu 22.04 system with a 3090, but from what I can tell it needs 9GB vram so a 3060 should do fine too.

'''
I created a new venv and then ran these commands to load transformers + other dependancies:

pip install torch==2.0.1+cu118 torchvision==0.15.2+cu118 torchaudio==2.0.2 --index-url https://download.pytorch.org/whl/cu118
pip install wheel 
pip install transformers accelerate flash-attn bitsandbytes
'''
import torch
from transformers import BitsAndBytesConfig, AutoModelForCausalLM, AutoTokenizer, pipeline

torch.random.manual_seed(0)
model_id = "microsoft/Phi-3-medium-128k-instruct"
quantization_config = BitsAndBytesConfig(llm_int8_threshold=200.0)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="cuda", 
    torch_dtype="auto", 
    trust_remote_code=True, 
    quantization_config=quantization_config,
    attn_implementation="flash_attention_2",
)
tokenizer = AutoTokenizer.from_pretrained(model_id)

messages = [
    {"role": "user", "content": "Can you provide ways to eat combinations of bananas and dragonfruits?"},
    {"role": "assistant", "content": "Sure! Here are some ways to eat bananas and dragonfruits together: 1. Banana and dragonfruit smoothie: Blend bananas and dragonfruits together with some milk and honey. 2. Banana and dragonfruit salad: Mix sliced bananas and dragonfruits together with some lemon juice and honey."},
    {"role": "user", "content": "What about solving an 2x + 3 = 7 equation?"},
]

pipe = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
)

generation_args = {
    "max_new_tokens": 500,
    "return_full_text": False,
    #"temperature": 0.0,
    "do_sample": False,
}

output = pipe(messages, **generation_args)
print(output[0]['generated_text'])