Source: https://seerepo.com/r/microsoft/unilm/built
Markdown source: https://seerepo.com/r/microsoft/unilm/built.md
Description: unilm is a Python repo built with FastAPI and Flask. It stores its data in pyarrow and webdataset. It has 769,203 written lines, with 163 test files.

# What is unilm built with?

unilm is a Python repo built with FastAPI and Flask. It stores its data in pyarrow and webdataset, uses AI through scikit-learn, torch, transformers, timm, torchvision, and 19 more, makes its content with pillow and lxml, is tested with Pytest, pytest-xdist, and seqeval, and deploys with boto3. It has 769,203 written lines, with 163 test files.

Read on November 5, 2026, at commit 31c5b90.

Repository: https://github.com/microsoft/unilm

## Frontend

### Interface

- **Gradio**: 3 files

## Backend

### The base

- **[FastAPI](https://seerepo.com/uses/fastapi)**
- **[Flask](https://seerepo.com/uses/flask)**: 1 file

### Data

- **[pyarrow](https://seerepo.com/uses/pyarrow)**: 16 files
- **webdataset**: 3 files

### AI

- **scikit-learn**
- **[torch](https://seerepo.com/uses/torch)**: 1,447 files
- **[transformers](https://seerepo.com/uses/transformers)**: 214 files
- **timm**: 56 files
- **torchvision**: 51 files
- **datasets**: 47 files
- **opencv-python**: 34 files
- **sacrebleu**: 25 files
- **[accelerate](https://seerepo.com/uses/accelerate)**: 20 files
- **sentencepiece**: 20 files
- **diffusers**: 19 files
- **deepspeed**: 15 files
- **torchaudio**: 13 files
- **[tokenizers](https://seerepo.com/uses/tokenizers)**: 10 files
- **bitsandbytes**: 8 files
- **einops**: 8 files
- **nltk**: 8 files
- **triton**: 8 files
- **[tiktoken](https://seerepo.com/uses/tiktoken)**: 7 files
- **vLLM**: 7 files
- **peft**: 2 files
- **torchmetrics**: 2 files
- **pycocotools**: 1 file

### Services

- **[OpenAI](https://seerepo.com/uses/openai)**: 2 files · [PFPO/data/openai_api_caller.py, line 6](https://github.com/microsoft/unilm/blob/31c5b904ca1bf2afb4c234a6675c683a4e5fc7cd/PFPO/data/openai_api_caller.py#L6)

## Support

### Content

- **[pillow](https://seerepo.com/uses/pillow)**: 60 files
- **[lxml](https://seerepo.com/uses/lxml)**: 2 files

### Testing

- **Pytest**
- **pytest-xdist**
- **seqeval**: 2 files

### Deploy

- **[boto3](https://seerepo.com/uses/boto3)**: 7 files

### Tools

- **antlr4-python3-runtime**
- **fire**
- **ipython**
- **jupyterlab**
- **mypy**
- **ninja**
- **Sphinx**
- **tensorboard**
- **packaging**: 18 files
- **tensorboardx**: 18 files
- **cython**: 5 files

### Other

- **absl-py**
- **cffi**
- **opencv-python-headless**
- **protobuf**
- **pybase64**
- **urllib3**
- **NumPy**: 524 files
- **omegaconf**: 243 files
- **tqdm**: 136 files
- **hydra-core**: 24 files
- **regex**: 17 files
- **SciPy**: 17 files
- **requests**: 16 files
- **wandb**: 14 files
- **pandas**: 13 files
- **pyyaml**: 9 files
- **termcolor**: 6 files
- **SymPy**: 5 files
- **attrs**: 3 files
- **ftfy**: 3 files
- **python-dateutil**: 2 files
- **filelock**: 1 file
- **natsort**: 1 file
- **networkx**: 1 file
- **shapely**: 1 file

## Other repos with a similar stack

- [unslothai/unsloth](https://seerepo.com/r/unslothai/unsloth/built): FastAPI, OpenAI, scikit-learn, torch, transformers, timm, datasets, accelerate, sentencepiece, diffusers, tokenizers, einops, tiktoken, peft, pillow, lxml · [Compare](https://seerepo.com/compare/microsoft/unilm/vs/unslothai/unsloth)
- [hiyouga/LlamaFactory](https://seerepo.com/r/hiyouga/LlamaFactory/built): FastAPI, Gradio, torch, transformers, torchvision, datasets, accelerate, sentencepiece, deepspeed, torchaudio, bitsandbytes, einops, nltk, tiktoken, peft · [Compare](https://seerepo.com/compare/hiyouga/LlamaFactory/vs/microsoft/unilm)
- [vllm-project/vllm](https://seerepo.com/r/vllm-project/vllm/built): FastAPI, OpenAI, torch, transformers, torchvision, datasets, sentencepiece, torchaudio, tokenizers, einops, triton, tiktoken, peft, pillow · [Compare](https://seerepo.com/compare/microsoft/unilm/vs/vllm-project/vllm)
- [datawhalechina/hello-agents](https://seerepo.com/r/datawhalechina/hello-agents/built): FastAPI, Flask, OpenAI, Gradio, scikit-learn, torch, transformers, datasets, accelerate, tokenizers, tiktoken, peft, pillow, lxml · [Compare](https://seerepo.com/compare/datawhalechina/hello-agents/vs/microsoft/unilm)
- [debpalash/VoiceStudio](https://seerepo.com/r/debpalash/VoiceStudio/built): FastAPI, pyarrow, OpenAI, Gradio, torch, transformers, torchvision, accelerate, diffusers, torchaudio, pillow, lxml · [Compare](https://seerepo.com/compare/debpalash/VoiceStudio/vs/microsoft/unilm)
- [comet-ml/opik](https://seerepo.com/r/comet-ml/opik/built): FastAPI, Flask, OpenAI, scikit-learn, torch, transformers, datasets, accelerate, tokenizers, tiktoken, peft, pillow · [Compare](https://seerepo.com/compare/comet-ml/opik/vs/microsoft/unilm)

## Frequently asked questions

### What language is unilm written in?

unilm is written in Python, with 769,203 written lines.

### What framework does unilm use?

unilm is built with FastAPI and Flask.

### What database does unilm use?

unilm stores its data in pyarrow and webdataset.

### Does unilm use AI?

Yes. unilm uses AI through scikit-learn, torch, transformers, timm, torchvision, datasets, opencv-python, sacrebleu, accelerate, sentencepiece, diffusers, deepspeed, torchaudio, tokenizers, bitsandbytes, einops, nltk, triton, tiktoken, vLLM, peft, torchmetrics, pycocotools, and OpenAI.

### Does unilm have tests?

Yes. unilm has 163 test files and is tested with Pytest, pytest-xdist, and seqeval.

### Which repos have a stack like unilm's?

The ones that share most of its stack: unslothai/unsloth, hiyouga/LlamaFactory, and vllm-project/vllm.

## What is your repo built with?

Any public repo, with no account: https://seerepo.com/r/owner/repo/built, with the owner and name it has on GitHub.