Skip to content

Latest commit

 

History

History
131 lines (92 loc) · 4.1 KB

File metadata and controls

131 lines (92 loc) · 4.1 KB

Using Phi-4-mini-mm to generate code

Phi-4-mini continues the strong coding capabilities of Phi Family. You can use Prompt to ask questions related to coding. Of course, after adding the strong reasoning ability, it has stronger coding capabilities, such as generating projects according to requirements.For example, generate projects according to requirements, such as:

Requirement

Create a Shopping Cart App

  • Create an API Rest with the following methods:
    • Get a list of beers using page offset and limit.
    • Get beer details by id.
    • Search for beer by name, description, tagline, food pairings, and price.
  • Create a list of products on the main page.
    • Create a search bar to filter products.
    • Navigate to the description page when the user clicks on a product.
  • (Optional) Slicer to filter products by price.
  • Create a shopping cart.
    • Add products to the cart.
    • Remove products from the cart.
    • Calculate the total price of the products in the cart.

Sample Code - Python

import requests
import torch
from PIL import Image
import soundfile
from transformers import AutoModelForCausalLM, AutoProcessor, GenerationConfig,pipeline,AutoTokenizer

model_path = 'Your Phi-4-mini-mm-instruct'

kwargs = {}
kwargs['torch_dtype'] = torch.bfloat16

processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True)

model = AutoModelForCausalLM.from_pretrained(
    model_path,
    trust_remote_code=True,
    torch_dtype='auto',
    _attn_implementation='flash_attention_2',
).cuda()

generation_config = GenerationConfig.from_pretrained(model_path, 'generation_config.json')

user_prompt = '<|user|>'
assistant_prompt = '<|assistant|>'
prompt_suffix = '<|end|>'

requirement = """

Create a Shopping Cart App

- Create an API Rest with the following methods:
    - Get a list of beers using page offset and limit.
    - Get beer details by id.
    - Search for beer by name, description, tagline, food pairings, and price.
- Create a list of products on the main page.
    - Create a search bar to filter products.
    - Navigate to the description page when the user clicks on a product.
- (Optional) Slicer to filter products by price.
- Create a shopping cart.
    - Add products to the cart.
    - Remove products from the cart.
    - Calculate the total price of the products in the cart."""

note = """ 

            Note:

            1. Use Python Flask to create a Repository pattern based on the following structure to generate the files

            |- models
            |- controllers
            |- repositories
            |- views

            2. For the view page, please use SPA + VueJS + TypeScript to build

            3. Firstly use markdown to output the generated project structure (including directories and files), and then generate the  file names and corresponding codes step by step, output like this 

               ## Project Structure

                    |- models
                        | - user.py
                    |- controllers
                        | - user_controller.py
                    |- repositories
                        | - user_repository.py
                    |- templates
                        | - index.html

               ## Backend
                 
                   #### `models/user.py`
                   ```python

                   ```
                   .......
               

               ## Frontend
                 
                   #### `templates/index.html`
                   ```html

                   ```
                   ......."""

prompt = f'{user_prompt}Please create a project with Python and Flask according to the following requirements:\n{requirement}{note}{prompt_suffix}{assistant_prompt}'

inputs = processor(prompt, images=None, return_tensors='pt').to('cuda:0')

generate_ids = model.generate(
    **inputs,
    max_new_tokens=2048,
    generation_config=generation_config,
)

generate_ids = generate_ids[:, inputs['input_ids'].shape[1] :]

response = processor.batch_decode(
    generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False
)[0]

print(response)