Ruliad (@ruliad) / X

Ruliad

20 posts

Ruliad

@ruliad

AI research lab

Earth

Joined June 2021

Ruliad reposted
Ruliad
@ruliad
Dec 4, 2024
Introducing DeepThought-8B: Transparent reasoning model built on LLaMA-3.1 with test-time compute scaling. - JSON-structured thought chains & controllable inference paths. - ~16GB VRAM, competitive w/ 70B models. - Open model weights, and inference scripts.
00:00
219K
Ruliad
@ruliad
Dec 4, 2024
Replying to @ruliad
The Deepthought model series unlock new capabilities and value propositions, with interesting research paths in the LLM paradigm in the short-term. Our research at Ruliad extends beyond current architectures and algorithms aiming to unlock the next leap in AI models. Join us in
3.4K
Ruliad
@ruliad
Dec 4, 2024
Replying to @ruliad
About Ruliad - our mission is self-sovereign general intelligence for all. We primarily work with organizations globally to inject private, secure and competent intelligence across applications and domains. Learn more at ruliad.co
00:00
4.2K
Ruliad
@ruliad
Dec 4, 2024
Replying to @ruliad
We are currently offering an API on a first-come first-served basis. Reach out to [email protected] to request access. We plan to eventually streamline this process with self-serve and self-sovereign deployment options in the coming weeks.
2.3K
Ruliad
@ruliad
Dec 4, 2024
Replying to @ruliad
This model is only the beginning with even more powerful versions in training. We look forward to any and all feedback you may have on where things can be improved so we can open source more advanced reasoning models for everyone. Give us feedback on the chat interface to help
00:00
2.4K
Ruliad
@ruliad
Dec 4, 2024
Replying to @ruliad
From our internal testing, this model is highly performant, comparable to 70B-scale models and uniquely-powerful on specific styles of benchmarks like instruction-following. As always, take benchmarks with a grain of salt and test the model on your specific downstream tasks.
2.9K
Ruliad
@ruliad
Dec 4, 2024
Replying to @ruliad
Scripted Injection: Define specific reasoning steps that the model will follow in sequence.
00:00
2.8K
Ruliad
@ruliad
Dec 4, 2024
Replying to @ruliad
Thought-routing Injection: If a specific thought type occurs, follow with another specific type.
00:00
3.8K
Ruliad
@ruliad
Dec 4, 2024
Replying to @ruliad
Max-routing Injection: Maximum number of thoughts w/ guaranteed ending step types.
00:00
3.2K
Ruliad
@ruliad
Dec 4, 2024
Replying to @ruliad
Injections are a way to customize and guide the thought process of Deepthought. We've implemented a few on our interface and API for you to test and discover novel ways to control models to do what you want.
3.7K
Ruliad
@ruliad
Dec 4, 2024
Replying to @ruliad
Beyond the open model weights, you can use our interface for free to interact with the alpha 8B model & its unique capabilities with thought injections. chat.ruliad.co
4.8K
Ruliad
@ruliad
Dec 4, 2024
Replying to @ruliad
Deepthought weights are downloadable on HF:
ruliad/deepthought-8b-llama-v0.01-alpha · Hugging Face
From huggingface.co
10K
Ruliad
@ruliad
Dec 4, 2024
Introducing DeepThought-8B: Transparent reasoning model built on LLaMA-3.1 with test-time compute scaling. - JSON-structured thought chains & controllable inference paths. - ~16GB VRAM, competitive w/ 70B models. - Open model weights, and inference scripts.
00:00
219K