## How C liberated the Memes Of Production

How C liberated the Memes Of Production.

by

Christoph Lohmann <20h@r-36.net>

## In the Beginning: Tensorflow

https://www.tensorflow.org/

* Required huge GPUs.
* Required much time.
* Was not easy to run everywhere.

## The Dark Times

* Big companies held back AI.
* Big companies told us, AI will always be only possible on their
  hardware.
* Big companies told us, only their way is our way.

## Then came the Llama.

https://www.theverge.com/2023/3/8/23629362/meta-ai-language-model-llama-leak-online-misuse

Facebook/Meta's LLama model was leaked.

Facebook/Meta decided to not take it down.

Facebook/Meta decided to finally open source their model and regularily
release it.

But it all was made for Python, like Tensorflow.

## Then came C++ Llama.

https://github.com/ggerganov/llama.cpp

Inference of Meta's LLaMA model (and others) in pure C/C++

https://github.com/ggerganov/whisper.cpp

High-performance inference of OpenAI's Whisper automatic speech
recognition (ASR) model.

## Then came C.

https://github.com/ggerganov/ggml
http://ggml.ai/

  The ggml way
    Minimal
      We like simplicity and aim to keep the codebase as small and as
      imple as possible

    Open Core
      The library and related projects are freely available
      under the MIT license. The development process is open
      and everyone is welcome to join. In the future we may
      choose to develop extensions that are licensed for
      commercial use

   Explore and have fun!
      We built ggml in the spirit of play. Contributors
      are encouraged to try crazy ideas, build wild
      demos, and push the edge of what’s possible

## ggml

    Written in C
    16-bit float support
    Integer quantization support (4-bit, 5-bit, 8-bit, etc.)
    Automatic differentiation
    ADAM and L-BFGS optimizers
    Optimized for Apple Silicon
    On x86 architectures utilizes AVX / AVX2 intrinsics
    On ppc64 architectures utilizes VSX intrinsics
    No third-party dependencies
    Zero memory allocations during runtime

## gguf

https://huggingface.co/docs/hub/gguf

Hugging Face Hub supports all file formats, but has built-in features for
GGUF format, a binary format that is optimized for quick loading and
saving of models, making it highly efficient for inference purposes. GGUF
is designed for use with GGML and other executors. GGUF was developed by
@ggerganov who is also the developer of llama.cpp, a popular C/C++ LLM
inference framework. Models initially developed in frameworks like
PyTorch can be converted to GGUF format for use with those engines.

## Easy frontends.

https://www.nomic.ai/gpt4all

Run your GPT local in no time, everywhere.

## Stable Diffusion Followed

Stable Difussion:
https://github.com/CompVis/stable-diffusion

A slow python inference for image generation.

https://github.com/leejet/stable-diffusion.cpp
> Plain C/C++ implementation based on ggml, working in the same way as llama.cpp

## The Revolution at Bitreich

annna, how do I $someprompt?

/home/annna/scm/ggml/build/bin/gpt-2 \
	-m models/gpt-2-1558M/ggml-model.bin \
	-t 8

## The Revolution at Bitreich

*::*::*)
	query=${text#**::} query=${query%%::*}
	annna-say -s "${server}" -c "${channel}" \
	"$(stable-diffusion-emoji "${query}")" &

stable-diffusion-emoji calls stable-diffusion

## The Revolution at Bitreich

sdbase="/br/ai/stable-diffusion.cpp"
sdcmd="${sdbase}/build/bin/sd -t 16 --steps 20 -s -1 -H 512 -W 512 \
         --cfg-scale 8.0"
sdmodel="${SD_MODEL:-${sdbase}/models/v2-1_768-nonema-pruned-ggml-model-f16.bin}"
sdnegative='nipples, busty, nude, lowres, (worst quality, low quality,
            normal quality:2), cropped, infant, blur, haze,
            tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of
            frame, extra limbs, disfigured, deformed, body out of frame, bad anatomy,
            watermark, signature, cut off, low contrast, underexposed, overexposed,
            bad art, beginner, amateur, distorted face, blurry, draft, grainy, nsfw'
outputf="${1}"
shift 1

${sdcmd} -m ${sdmodel} -o "${outputf}" -p "$@" -n "${sdnegative}" \
	>/dev/null 2>&1

## Generated Meme Count

                annna meme count (total on 2024-08-03: 22834) set log y
   100000 +-----------------------------------------------------------------+
          |+         +          +          +          +          +         +|
          |+ total ***T***                                         TTTTTT  +|
          |+ memes ###M###                 TMMMMMMMMMMMMMMMMMMMMMMMMMMMMM  +|
    10000 |-emojis $$$E$$$      MMMMMMMMMMMMM                             +-|
          |filters %%%F%%% MMMMMM                                          +|
          |+             MMM                                            E  +|
     1000 |-+          MMM                              FFFFFFFFFFFFFFFFF +-|
          |+          MM                      FFFFFFFFFFF       EEE        +|
          |+     MMMMMM                   FFFFF                 E          +|
          |+  MMMM                       FF                    EE          +|
      100 |-MMM                        FFF                     E          +-|
          |MM                          F                       E           +|
          |M                          FF                       $           +|
       10 |M+                         F                        E          +-|
          |#                                                               +|
          |M                                                               +|
          |+         +          +          +          +          +         +|
        1 +-----------------------------------------------------------------+
       2018-12    2019-12    2020-12    2021-12    2022-12    2023-12   2024-12

## The Revolution at Bitreich

And that is just the beginning.

When resources get cheaper, we can do:
	* Image modification
	* Image changing
	* Text2Video
	* Video modification

Want to help?

## Thanks

Thanks for Listening.

Questions?