dispatch
transformers
dispatch | transformers | |
---|---|---|
20 | 176 | |
4,602 | 125,021 | |
1.0% | 1.4% | |
9.9 | 10.0 | |
4 days ago | 7 days ago | |
Python | Python | |
Apache License 2.0 | Apache License 2.0 |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
dispatch
- Netflix Dispatch
-
Is there any open source project that uses FasAPI?
They use only sync routes in the project and can’t explain why https://github.com/Netflix/dispatch/issues/1073
- Is it really advisable to try to run fastapi with predominantly sync routes in a real world application?
-
How to build a scalable project file structure for a beginner.
By far my favorite production FastAPI app to use as a references of how to use these technologies well is NetFlix Dispatch: https://github.com/Netflix/dispatch
-
FastAPI Boilerplate using MongoDB, Motor, Docker
Hey, I have a lot of opinions about this template, but these are just my opinions based on my own experiences being burned by these things so take from them what you will: 1. Your version of poetry is outdated, dependency groups don't work that way anymore and this will fail to install on modern poetry 2. You list pyyaml as a dependency but don't use it anywhere 3. The healthcheck endpoint is interesting, but expensive and a security risk. I like the value this provides, but I don't know if exposing it this way or using it as a healthcheck is a good idea 1. You typically don't want to touch external systems (mongo) as part of a healthcheck as this can cause cascading failure chains that get out of hand quickly 2. You typically don't want to touch the underlying system itself 1. which means you can / should get rid of psutil as a dependency 4. You don't need and shouldn't use pytest-asyncio for a FastAPI project. It comes built-in with its own async test handlers that you should be using 5. Having python-dotenv installed in production has burned me many times. I recommend removing this complete, otherwise just moving it to a dev dep 6. Using the src layout prevents a lot of weird import time problems from cropping up in production, I recommend checking it out 7. The entrypoint for the Docker container should be using 1 worker, as containers really prefer if you have only a single root PID chain and nothing else. Deploying this into k8s would cause a lot of issues 8. Native python logging really isn't great for modern production applications. Structlog or Loguru are great alternatives and much easier to use (which should remove your only dependency on pyyaml) 9. The configuration management may not work the way you want since it is weakly typed. Since FastAPI uses Pydantic, you have access to BaseSettings which is a far superior product for configuration management, especially with environment variables 10. The app and API folder structure is an anti-pattern that doesn't scale past projects the size of a tutorial on how to laern FastAPI. I strongly recommend changing this to move of a vertical slice or folder per feature layout such as is used in https://github.com/Netflix/dispatch/tree/master/src/dispatch 11. FastAPI routes don't need `response_model=` anymore in favor of adding the return type to your function signature such as `async def create_thing() -> Thing:` 12. The uuid_masker function is interesting, but exposing UUIDs in logs usually doesn't pose a security risk and only makes debugging more difficult 13. You have some type lies in your code that could burn you such as https://github.com/alexk1919/fastapi-motor-mongo-template/blob/main/app/db/db.py#L10 . This pattern for the global DB handle has also burned me in the past and I had to go back and refactor out all of them to instead to purely use the FastAPI dependency injection chaining 14. datetime.datetime isn't safe to use as it is in sample_resource_common.py, you need a timezone aware implementation 15. Your test suite is stateful, require a running database, leak a lot of implementation details of the underlying models. This is every anti-pattern in the book for unit testing. And if you are going to do integration tests, then you would be better off with tooling designed for it such as playwright. Again, these are all just my opinions and may alone not be enough to warrant changing anything you have here.
-
Python projects with best practices on Github?
Two random examples I found from 30 seconds of googling: Here’s Netflix using it in their crisis management tool, and here’s Uber using it in their deep learning framework.
-
Open Source Projects based on FastAPI
netflix dispatch
-
As a long time programmer what are some important coding styles ?
As someone who uses FastAPI, I find the https://github.com/Netflix/dispatch code to be a great reference.
-
CEO faces backlash after quoting Martin Luther King Jr. in announcing layoffs
Besides that paying $21 to $41 per user for this nuts. Set up a VPS with Dispatch (opensourced by Netflix) and save your company some money.
-
Total beginner, use FastAPI?
For production ready code examples I use: https://github.com/Netflix/dispatch
transformers
-
AI enthusiasm #9 - A multilingual chatbot📣🈸
transformers is a package by Hugging Face, that helps you interact with models on HF Hub (GitHub)
-
Maxtext: A simple, performant and scalable Jax LLM
Is t5x an encoder/decoder architecture?
Some more general options.
The Flax ecosystem
https://github.com/google/flax?tab=readme-ov-file
or dm-haiku
https://github.com/google-deepmind/dm-haiku
were some of the best developed communities in the Jax AI field
Perhaps the “trax” repo? https://github.com/google/trax
Some HF examples https://github.com/huggingface/transformers/tree/main/exampl...
Sadly it seems much of the work is proprietary these days, but one example could be Grok-1, if you customize the details. https://github.com/xai-org/grok-1/blob/main/run.py
-
Lossless Acceleration of LLM via Adaptive N-Gram Parallel Decoding
The HuggingFace transformers library already has support for a similar method called prompt lookup decoding that uses the existing context to generate an ngram model: https://github.com/huggingface/transformers/issues/27722
I don't think it would be that hard to switch it out for a pretrained ngram model.
-
AI enthusiasm #6 - Finetune any LLM you want💡
Most of this tutorial is based on Hugging Face course about Transformers and on Niels Rogge's Transformers tutorials: make sure to check their work and give them a star on GitHub, if you please ❤️
-
Schedule-Free Learning – A New Way to Train
* Superconvergence + LR range finder + Fast AI's Ranger21 optimizer was the goto optimizer for CNNs, and worked fabulously well, but on transformers, the learning rate range finder sadi 1e-3 was the best, whilst 1e-5 was better. However, the 1 cycle learning rate stuck. https://github.com/huggingface/transformers/issues/16013
-
Gemma doesn't suck anymore – 8 bug fixes
Thanks! :) I'm pushing them into transformers, pytorch-gemma and collabing with the Gemma team to resolve all the issues :)
The RoPE fix should already be in transformers 4.38.2: https://github.com/huggingface/transformers/pull/29285
My main PR for transformers which fixes most of the issues (some still left): https://github.com/huggingface/transformers/pull/29402
- HuggingFace Transformers: Qwen2
- HuggingFace Transformers Release v4.36: Mixtral, Llava/BakLlava, SeamlessM4T v2
- HuggingFace: Support for the Mixtral Moe
-
Paris-Based Startup and OpenAI Competitor Mistral AI Valued at $2B
If you want to tinker with the architecture Hugging Face has a FOSS implementation in transformers: https://github.com/huggingface/transformers/blob/main/src/tr...
If you want to reproduce the training pipeline, you couldn't do that even if you wanted to because you don't have access to thousands of A100s.
What are some alternatives?
fastapi-best-practices - FastAPI Best Practices and Conventions we used at our startup
fairseq - Facebook AI Research Sequence-to-Sequence Toolkit written in Python.
fastapi - FastAPI framework, high performance, easy to learn, fast to code, ready for production
sentence-transformers - Multilingual Sentence & Image Embeddings with BERT
full-stack-fastapi-template - Full stack, modern web application template. Using FastAPI, React, SQLModel, PostgreSQL, Docker, GitHub Actions, automatic HTTPS and more.
llama - Inference code for Llama models
fastapi-router-controller - A FastAPI utility to allow Controller Class usage
transformer-pytorch - Transformer: PyTorch Implementation of "Attention Is All You Need"
opal - Fork of https://github.com/permitio/opal
text-generation-webui - A Gradio web UI for Large Language Models. Supports transformers, GPTQ, AWQ, EXL2, llama.cpp (GGUF), Llama models.
databases - Async database support for Python. 🗄
huggingface_hub - The official Python client for the Huggingface Hub.