LAPIS

model-to-procslib

2 min read

Integrate existing model inference codes into procslib.

Assuming full codebase is given, gpt-o1 usually oneshots these problems and only needs minor tweaks for outputs formatting

Prompt Template:

I have a disorganized repository with some Python files, a pyproject.toml, and random data/checkpoint directories. I want to:

  1. Make it a standard Python package named my_project, using the src/my_project layout. So the final structure is something like:
my_project/ 
├─ pyproject.toml 
├─ requirements.txt 
├─ src/ 
│ └─ my_project/ 
│ ├─ **init**.py 
│ ├─ main_inference_class.py 
│ ├─ ... 
└─ ...
  1. Remove any local path hacks (sys.path.append(...), etc.) and switch them to either:
from .some_module import ...

or

from my_project.some_module import ...

so that the code runs after pip install ..

  1. Convert hard-coded local checkpoint paths into code that fetches from Hugging Face via huggingface_hub.hf_hub_download. For example:

    from huggingface_hub import hf_hub_download
 
local_ckpt_path = hf_hub_download("some_user/my_model_repo", "my_checkpoint.pt")

  2. Create a single Inference class that loads the model (downloading from Hugging Face if no local path is provided), plus a demonstration snippet that uses it. Example:

from my_project.main_inference_class import MyInference

Instantiate

infer = MyInference( model_repo=“some_user/my_model_repo”, model_filename=“my_checkpoint.pt”, )

Inference

audio = infer(“Hello world”)


3. **Update `pyproject.toml`** so that it includes any needed dependencies in `[project.dependencies]` and also includes a `[project.scripts]` entry if we want a CLI. Example:

```toml
[project]
name = "my_project"
version = "0.1.0"
description = "..."
requires-python = ">=3.10"
dependencies = [
    "torch",
    "huggingface_hub",
    "soundfile",
    ...
]
[project.scripts]
my_project-cli = "my_project.__main__:main"
  1. Demonstrate that the user can install with pip install . and then run the inference code or the CLI command.

Given these steps, please provide a concise diff or snippet for each file so we can see exactly what needs to be changed. The final result should be a neatly installable Python package.

code dump:

<YOUR-CONTENT-HERE>