Type

new chapter

Chapter/Page

Something else

Description

No response

Follow-up to #42

• should we have :valid and :invalid CSS?
• also can hitting the Enter key work (for keyboard warriors who don't like clicking buttons)?

Have had reports of people giving up on accessing the book because they didn't realise they currently have to use the mouse to click Submit

Licences — State of Open Source AI Book

https://book.premai.io/state-of-open-source-ai/licences/

Type

new chapter

Chapter/Page

Something else

Description

Right now RAG is written under Fine-tuning section. However, I am not very much sure, whether that would be the correct place. Since, RAG does not change any parameters of the model or adds new parameters.

So, it would be much better to add it under different section or a dedicated chapter or add it to the VectorDatabase chapter. The concept of RAG is actually not very new. FAISS was used readily before the surge if LLMs (mostly in image search or semantic search).

In the RAG chapter we can populate by these contents:

1. What is RAG (already covered).
2. How RAG works on a high level.
3. RAG in LLMs.
4. RAG/vector dbs in computer vision.

Type

new chapter

Chapter/Page

fine-tuning

Description

Similar to LLMs, we use similar fine-tuning strategies for diffusion models too. So this chapter should include different ways of fine-tuning those models.

I'm not aware of any licence that works for both code & text. I'd suggest separate licences for each (this is quite common elsewhere):

• Text: CC-BY-4.0
• Code: Apache-2.0

thoughts @filopedraz?

Type

other (e.g. typos, factual errors, etc.)

Chapter/Page

Something else

Description

Description

We need to go back to the email wall, but we need to make sure that email provided are valid.

• initial config #7
• CI build & deploy #9
  • basic CI linting
• DevContainer #8
• contrib guide (syntax, style) #10

Software Development toolKits — State of Open Source AI Book

https://book.premai.io/state-of-open-source-ai/sdk/

Type

new chapter

Chapter/Page

Something else

Description

No response

Type

new chapter

Chapter/Page

Something else

Description

No response

https://github.com/premAI-io/state-of-open-source-ai/issues

it's difficult to keep track of all the innovations. There's been enormous progress in the field in {term}the last year .

this links to sota? why that it doesn't make sense

while following state-of-open-source-ai/#editing-the-book, it breaks on line jupyter-book build -b dirhtml --all . with the following error:

➜ jupyter-book build -b dirhtml --all .
Usage: jupyter-book build [OPTIONS] PATH_SOURCE
Try 'jupyter-book build -h' for help.

Error: No such option: -b

Have you considered adding a chapter dedicated to security of open source AI? Can help contribute.

Use markdown-MyST syntax (e.g. source ➡️ rendered)
Please also report any jupyter-book build framework-related problems/questions in #12 :)

Type

other (e.g. typos, factual errors, etc.)

Chapter/Page

fine-tuning

Description

The fine-tuning chapter should include all the tools and frameworks in order to do fine-tuning.

Type

new chapter

Chapter/Page

Something else

Description

No response

Type

other (e.g. typos, factual errors, etc.)

Chapter/Page

Something else

Description

example

Unrendered trailing character

No trailing character shown

Ascetic issue, not serious

Type

new URL/reference/table row

Chapter/Page

fine-tuning

Description

Expand the Finetuning chapter in order to include the DRLX technique.

Type

new chapter

Chapter/Page

Something else

Description

Description

• https://github.com/IBM/ModuleFormer
• https://github.com/SkunkworksAI/hydra-moe
• Decentralized Mixture of Experts

Type

new chapter

Chapter/Page

eval-datasets

Description

The evaluation page is really good, however, it would be awesome if we could add some information on the following evaluation frameworks.

1. HELM by Stanford.
2. LM Evaluation Harness by Eluther AI.
3. Code Evaluation Harness by BigCode.

The content should be mainly regarding how they are trying to do evaluation and how to get started with each.

• readme: make readable on GitHub (split bits into a separate root: index.md) #13
  • html: use dirhtml builder #13
• tests: add linkchecks #13
• html: fix comments integration #13
• html: create remaining chapter stubs #14
• assets: use an external static image host to avoid bloating this repo: static.premai.io #17
• credit GH contributors (avatars + links) #17
• html: Bibliography borkes right navbar. Find a fix or report upstream issue? #17
  • https://sourceforge.net/p/docutils/patches/195
  • executablebooks/jupyter-book#2048
• add PR review template & improve contrib guide #17
• fix concurrent builds #17
• fix build instructions #21
• misc minor improvements #23
• pdf: improve formatting
  • replace the existing page download (print to pdf) links with proper pdf #86
  • improve cover pages #65
  • fix cross-ref linking (e.g. links to chapters are not clickable) #79
  • substitute some unicode chars (e.g. 🟢, 🟡, 🔴, https://tex.stackexchange.com/questions/601479) #79
• html: improve style
  • #42
  • #84 #83 #77 #70 #63
• deploy: pick a new subdomain e.g. book.premai.io #24
  • update repo_url & references #24
  • plausible #25
• contrib list per-chapter #36
  • https://github.com/casperdcl/git-fame#codeowners
  • https://stackoverflow.com/questions/44888187/get-github-username-through-primary-email
• contrib: Zenodo integration config (CITATION.cff/.zenodo.json) #41
• open links in new tab #55 <- #50
  • executablebooks/jupyter-book#1070
  • https://stackoverflow.com/questions/25583581/add-open-in-new-tab-links-in-sphinx-restructuredtext
  • sphinx-doc/sphinx#2550
• convert links to references where appropriate (e.g. arxiv)
• make repo public
  • enables features like utteranc.es comments, CI status badges, dependabot, external PRs, GH contributor lists, Zenodo DOIs, etc.
  • fix Zenodo #92
  • fix utterances #81
  • tag v1
• add https://ogp.me metadata #86
• add to https://github.com/executablebooks/meta/blob/main/docs/gallery.yml executablebooks/meta#1090

Follow-up to #11

v1.1

• rename repo?
• fix all TODO notes
• fix all {{ wip_chapter }}/% TODO/Work in Progress notes
• CI: speed up builds
  • maybe parallel linkcheck + html + pdf #69
  • smaller pdf docker container? (easier since https://github.com/xu-cheng/latex-action/releases/tag/3.0.0) (alternatively, use different/quicker pdf action?)
  • add minimal pdf LaTeX to DevContainer (and use https://github.com/devcontainers/ci)?

• create model-formats.md (see e.g. 70892bd for adding a chapter)

Use markdown-MyST syntax (e.g. source ➡️ rendered)

Potential subheadings:

• GGML
• ONNX
• TVM

Please also report any jupyter-book build framework-related problems/questions in #12 :)

Evaluation & Datasets — State of Open Source AI Book

https://book.premai.io/state-of-open-source-ai/eval-datasets/

Type

other (e.g. typos, factual errors, etc.)

Chapter/Page

Something else

Description

Use new Prem logo and guidelines.

Type

new chapter

Chapter/Page

fine-tuning

Description

The contents on this chapter should be the following:

1. What is Parameter efficient fine-tuning methods and why we do it.
2. Different PEFT methods:
   a. Prompt Tuning
   b. LoRA
   c. QLoRA
   etc

Type

new chapter

Chapter/Page

Something else

Description

No response

Type

new chapter

Chapter/Page

Something else

Description

I suggest that we have a separate chapter on Quantization. The chapter should include all the new methods created to handle quantization and the most famous libraries associated with them.

• GPTQ
• AWQ

https://github.com/premAI-io/state-of-open-source-ai/blob/main/desktop-apps.md

Use markdown-MyST syntax (e.g. source ➡️ rendered)
Please also report any jupyter-book build framework-related problems/questions in #12 :)

Type

other (e.g. typos, factual errors, etc.)

Chapter/Page

Something else

Description

For many people the most effective way to read a book is in audiobook format.
I would like to ask to have some text-to-speech transcription available, or, considering the involved costs, a text-to-speech friendly artifact available, so I can execute the TTS by myself.
I volunteer to provide the generated "audiobook" to be shared among this book audience.

Type

new chapter

Chapter/Page

Something else

Description

No response

Type

new chapter

Chapter/Page

Something else

Description

Doing training or inference models are fairly easy, when we have smaller number of parameters. But when the scale of parameters and data increases, it becomes increasingly difficult to optimize that (interms of compute and performance). So a dedicated chapter on inference optimization and arithmetics on resource calculations becomes very useful.

Some reference links:

• Mastering LLM Inference by NVIDIA
• Transformer Math 101
• Transformer Inference Arithmetic
• A twitter thread on some optimizations

Type

new chapter

Chapter/Page

models

Description

Although MoE has been there in general Deep learning, just now Mistral Released their second models with MoE, so now it is very important to release those also.

When signing in with GitHub to add a comment in a chapter I've got 404.

Models — State of Open Source AI Book

https://book.premai.io/state-of-open-source-ai/models/

• create mlops-engines.md (see e.g. 70892bd for adding a chapter)

Use markdown-MyST syntax (e.g. source ➡️ rendered)

Potential subheadings:

• Difficulties of Working with OpenSource MLOps
• Python Bindings and More
• PyTorch Toolchain - From C/C++ to Python
• llama.cpp
• ONNX Runtime
• Apache TVM

Please also report any jupyter-book build framework-related problems/questions in #12 :)

Hey Team,

As you know, we've been trying to incorporate a custom index.html page in our [Jupyter Book/Sphinx] book that does not adhere to the standard theme and I've hit a roadblock.

Challenge

After exploring various options, it appears that Jupyter Book and Sphinx are not designed to easily allow a completely custom index.html that diverges from the set themes so the https://premai-book.netlify.app seems unusable at this point.

Proposed Solution

The fastest way to implement our custom index.html seems to be post-build where we'd:

1. Run the standard Jupyter Book or Sphinx build command to generate the documentation.
2. Execute a post-build script to replace the default index.html with our custom version.

Next Steps

• I wouldn't want to break anything with the current site so would need someone to help with this script