Constantinus

@constantinus

737 Following

203 Followers

Constantinus pfp

What AI agents inside warpcast do you know? Please links.

0 reply

0 recast

0 reaction

Constantinus pfp

It's interesting to watch the trend of onchain AI agents that are valued in the tens and hundreds of $M. Realizing that many of them have no technology at all except APIs and function colling and RAG.

0 reply

0 recast

0 reaction

Constantinus pfp

TIPO (Text to Image with text presampling for Prompt Optimization) is a technique that improves the quality and usability of text-2-image models. TIPO uses LLM to preprocess text queries, making them more accurate and informative. It accepts both natural language Prompts and Danbooru tag format. The basic idea behind the method is that more detailed and specific queries lead to more accurate image generation, while unspecific queries lead to a wider range but less accurate results. TIPO generates multiple detailed variants of a query from one simple query, thereby expanding the space of possible results and increasing the probability of obtaining the desired image. We present 2 models of TIPO, both built on LLaMA 400M, trained on Danbooru2023, GBC10M and Coyo-HD-11M sets with a total of 30 mlrn tokens. https://github.com/KohakuBlueleaf/KGen

0 reply

0 recast

0 reaction

Constantinus pfp

SmolLM2: the second generation of compact LLMs from HuggingFace. Hugging Face introduced SmolLM2, a new series of SLMs optimized for resource-constrained devices and designed for English text generation, summarization and function calling tasks. SmolLM2 models were trained on a mix of FineWeb-Edu, DCLM and Stack datasets. Post-training testing showed the superiority of the older SmolLM2-1.7B model over Meta Llama 3.2 1B and Qwen2.5-1.5B. The models are available in three configurations: 135M, 360M and 1.7B parameters, each model has its own Instruct-version, and the 1.7B and 360M are also official GGUF quantized versions: SmolLM2-1.7B https://huggingface.co/HuggingFaceTB/SmolLM2-1.7B SmolLM2-360M https://huggingface.co/HuggingFaceTB/SmolLM2-360M SmolLM2-135M https://huggingface.co/HuggingFaceTB/SmolLM2-135M

0 reply

0 recast

0 reaction

Constantinus pfp

A project on Github that offers to run LLM on AMD graphics gas pedals using the Docker container. The image is designed to work with Hugging Face models, primarily the LLama family of models. To run it, you need to have an AMD GPU with ROCm support (version 5.4.2 or higher) and Docker installed. To adapt the inference logic to your needs, make appropriate changes to the run_inference.py file followed by rebuilding the Docker image. The project provides an Aptfile file containing a list of required ROCm packages (rocm-dev, rocm-libs, rocm-cmake, miopen-hip and rocblas) to be installed in the Docker container. https://github.com/slashml/amd_inference

0 reply

0 recast

1 reaction

Constantinus pfp

ExVideo is a tuning technique to improve the ability of models to generate video ExVideo allows a model to generate 5x more frames while requiring only 1.5k GPU training hours on a dataset of 40k videos. https://ecnu-cilab.github.io/ExVideoProjectPage/ https://huggingface.co/papers/2406.14130

0 reply

0 recast

1 reaction

Constantinus pfp

GaussianObject: Just Taking Four Images to Get A High-Quality 3D Object with Gaussian Splatting. ▪Github: https://github.com/GaussianObject/GaussianObject ▪Colab: https://colab.research.google.com/drive/1WIZgM--tJ3aq25t9g238JAuAoXrQYVMs?usp=sharing#scrollTo=TlrxF62GNePB ▪Project: https://gaussianobject.github.io

0 reply

0 recast

1 reaction

Constantinus pfp

It's a great research for a Sunday evening. https://paperswithcode.com/paper/storydiffusion-consistent-self-attention-for

0 reply

0 recast

0 reaction

Constantinus pfp

The news is encouraging. I think I can use it in my projects. https://warpcast.com/dwr.eth/0xad16b85a

1 reply

0 recast

1 reaction

Constantinus pfp

I don't have time to test new models. The increasing speed of AI development is amazing.

0 reply

0 recast

0 reaction

Constantinus pfp

Gm guys. Listen to this. https://suno.com/song/f11f0e7d-32e0-480c-8fdf-1f453d96cb16

2 replies

0 recast

4 reactions

Constantinus pfp

I'm convinced that /onchain is a team game. Many players are already teaming up to find secrets. I present you the concept of a subGuild . Guild within an /onchain Metaguild where people unite by interest, earn points together, share artifacts with the help of tools like Gnosis Safe, Delegate.xyz, PartyDAO.

5 replies

2 recasts

14 reactions

Constantinus pfp

0 reply

0 recast

3 reactions

Constantinus pfp

It'll be even more affordable soon. https://arxiv.org/abs/2404.07413

0 reply

0 recast

2 reactions

Constantinus pfp

Gm. I have questions...

0 reply

0 recast

5 reactions

Constantinus pfp

Audience attention is the hardest thing to keep in crypto. What tools are available now? - Economic incentive. - Gamification. What's the next tool? I'm convinced adaptability.

0 reply

0 recast

3 reactions

Constantinus pfp

Hey, guys! An error appears when creating a guild: This telegram reward belongs to a different guild. IDb5d44779-45e4-4e10-b3c1-5c0dee937a73 Guild bot added to telegram group with administrator rights. Telegram group is not used in any guilds. Can you please tell me the error in my actions or in the frontend?

0 reply

0 recast

4 reactions

Constantinus pfp

0 reply

0 recast

1 reaction

Constantinus pfp

Since 2021 I have been passionate about onchain reputation. Back then I proposed to many communities to create an end-to-end token-based onchain reputation system, but nobody supported me then. Now everything has changed, it seems that the time has come.

0 reply

0 recast

1 reaction

Constantinus pfp

Overall, I like the dynamic that @raz is creating in /onchain. It feels like it doesn't always go according to plan, but it also feels like a lot of work has been done. Do you think that what happens is a well thought out script or improvisation? I think it's a symbiosis.

1 reply

0 recast

6 reactions