85 9 185

t.d.a.g. PRO

sequelbox

sequelbox.bsky.social

AI & ML interests

open source, infinite games. (they/them)

Recent Activity

updated a dataset about 2 hours ago

sequelbox/Celestia3-DeepSeek-R1-0528-PREVIEW

new activity about 2 hours ago

sequelbox/Celestia3-DeepSeek-R1-0528-PREVIEW:[bot] Conversion to Parquet

posted an update 1 day ago

EARLY SNEAK PREVIEW: get a first look at the Celestia 3 science-reasoning dataset, built with DeepSeek's newest R1-0528 reasoning model! Subjects include physics, chemistry, biology, computer science, Earth science, astronomy, and information theory. This early look contains the first 14k rows, all synthetic responses using https://huggingface.co/deepseek-ai/DeepSeek-R1-0528 SEE IT HERE: https://huggingface.co/datasets/sequelbox/Celestia3-DeepSeek-R1-0528-PREVIEW Support our releases: https://huggingface.co/spaces/sequelbox/SupportOpenSource Coming up we'll have more dataset releases, including some novel reasoning and analysis methods - we think an important role for open source researchers is experimenting with new response styles on top of the increasingly excellent base models available to finetune. more to come soon! allegra

View all activity

Organizations

sequelbox's activity

posted an update 1 day ago

Post

297

EARLY SNEAK PREVIEW: get a first look at the Celestia 3 science-reasoning dataset, built with DeepSeek's newest R1-0528 reasoning model! Subjects include physics, chemistry, biology, computer science, Earth science, astronomy, and information theory.

This early look contains the first 14k rows, all synthetic responses using deepseek-ai/DeepSeek-R1-0528

SEE IT HERE: sequelbox/Celestia3-DeepSeek-R1-0528-PREVIEW

Support our releases: sequelbox/SupportOpenSource

Coming up we'll have more dataset releases, including some novel reasoning and analysis methods - we think an important role for open source researchers is experimenting with new response styles on top of the increasingly excellent base models available to finetune.

more to come soon!
allegra

posted an update 6 days ago

Post

282

NEW RELEASE: we've brought Esper 3 to the new deepseek-ai/DeepSeek-R1-0528-Qwen3-8B model!

- A full-stack software assistant: a reasoning finetune focused on coding, architecture, and DevOps using the Titanium and Tachibana datasets!
- Improved general and creative reasoning skills, powered by the Raiden dataset.

Get the newest Esper 3: ValiantLabs/DeepSeek-R1-0528-Qwen3-8B-Esper3
Support our releases: sequelbox/SupportOpenSource

more on the way next week!

celestially yours ;)
allegra

replied to their post 7 days ago

we'll be expanding Qwen sizes in both directions :) thanks for your review!

posted an update 8 days ago

Post

321

Updates for the week:
- released some new merge models using ValiantLabs/Qwen3-14B-Esper3 and other Qwen 3 14b finetunes - these merges include math, Web3, uncensored, and general mix. depending on your use case for Esper 3 these may be helpful to you! find them at @sequelbox
- coming up we'll have more model sizes for Esper 3 and Cobalt 2, releasing soon!
- also super excited for more dataset releases with the newly released deepseek-ai/DeepSeek-R1-0528

Support the above efforts and others: sequelbox/SupportOpenSource

back to building :)

2 replies

reacted to lukmanaj's post with 👍 8 days ago

Post

2343

I am so happy to share to all that I’ve just completed the first unit of the new MCP course on Hugging Face and earned my certificate! The AI acceleration track is intense and fast-paced, but I’m doing my best to keep up. Excited for what’s ahead!

1 reply

posted an update 17 days ago

Post

1873

NEW RELEASE: Cobalt 2 for Qwen 3 14b!

- A math-reasoning finetune, focused on high-difficulty math questions with the zwhe99/DeepMath-103K dataset!
- Improved general and creative reasoning skills, powered by the Raiden dataset

GET IT NOW: ValiantLabs/Qwen3-14B-Cobalt2
HELP US RELEASE FASTER: sequelbox/SupportOpenSource

we've got more releases to come soon - excited to share with everyone!

love,
allegra

posted an update 21 days ago

Post

1283

Esper 3 is now available for Qwen 3 14b!

- A full-stack software assistant: a reasoning finetune focused on coding, architecture, and DevOps using the Titanium and Tachibana datasets!
- Improved general and creative reasoning skills, powered by the Raiden dataset.

GET IT NOW: ValiantLabs/Qwen3-14B-Esper3
HELP US RELEASE FASTER: sequelbox/SupportOpenSource

more to come :)
allegra

posted an update about 1 month ago

Post

2689

NEW RELEASE: Esper 3 for Qwen 3!

- A full-stack software assistant: a reasoning finetune focused on coding, architecture, and DevOps using the Titanium and Tachibana datasets!
- Improved general and creative reasoning skills, powered by the Raiden dataset.

4B model: ValiantLabs/Qwen3-4B-Esper3
8B model: ValiantLabs/Qwen3-8B-Esper3

We'll also be bringing Esper 3 to larger Qwen 3 models as soon as we can - if you want these, consider helping us out: sequelbox/SupportOpenSource

More models and datasets to come soon!

with my love and enthusiasm,
allegra

posted an update about 2 months ago

Post

1787

TITANIUM 2 Deepseek-R1 dataset is here! Open-source synthetic architecture and DevOps dataset: sequelbox/Titanium2-DeepSeek-R1

Esper 3 will be coming out soon for multiple base models, trained on Titanium, Raiden, and more :)

with my love,
allegra

reacted to mkurman's post with ❤️ 3 months ago

Post

3705

Introducing a new architecture, MedIT One – a single-token transformer with LSTM-like recurrence.

It is extremely fast in training and inference, but we lack funding for large-scale training. Enjoy 🍓

https://github.com/MedITSolutionsKurman/medit-one

reacted to singhsidhukuldeep's post with 👍 3 months ago

Post

6937

Exciting New Tool for Knowledge Graph Extraction from Plain Text!

I just came across a groundbreaking new tool called KGGen that's solving a major challenge in the AI world - the scarcity of high-quality knowledge graph data.

KGGen is an open-source Python package that leverages language models to extract knowledge graphs (KGs) from plain text. What makes it special is its innovative approach to clustering related entities, which significantly reduces sparsity in the extracted KGs.

The technical approach is fascinating:

1. KGGen uses a multi-stage process involving an LLM (GPT-4o in their implementation) to extract entities and relations from source text
2. It aggregates graphs across sources to reduce redundancy
3. Most importantly, it applies iterative LM-based clustering to refine the raw graph

The clustering stage is particularly innovative - it identifies which nodes and edges refer to the same underlying entities or concepts. This normalizes variations in tense, plurality, stemming, and capitalization (e.g., "labors" clustered with "labor").

The researchers from Stanford and University of Toronto also introduced MINE (Measure of Information in Nodes and Edges), the first benchmark for evaluating KG extractors. When tested against existing methods like OpenIE and GraphRAG, KGGen outperformed them by up to 18%.

For anyone working with knowledge graphs, RAG systems, or KG embeddings, this tool addresses the fundamental challenge of data scarcity that's been holding back progress in graph-based foundation models.

The package is available via pip install kg-gen, making it accessible to everyone. This could be a game-changer for knowledge graph applications!

posted an update 4 months ago

Post

2827

Raiden is here! 63k creative-reasoning and analytic-reasoning prompts answered by DeepSeek's 685b R1 model!

- All prompts from microsoft/orca-agentinstruct-1M-v1 and all responses from deepseek-ai/DeepSeek-R1
- A deep look at R1's reasoning skills! Use as you will.

Get it now: sequelbox/Raiden-DeepSeek-R1

for everyone :)

reacted to rubenroy's post with 🚀 4 months ago

Post

2612

🔥🚀 Hey everyone! I'm excited to share my latest LLM release: Gilgamesh 72B, a model built on Qwen 2.5-72B Instruct. Gilgamesh was trained on a couple of my GammaCorpus datasets, specifically:

- rubenroy/GammaCorpus-CoT-Math-170k
- rubenroy/GammaCorpus-v2-5m
- rubenroy/GammaCorpus-Fact-QA-450k

I've submitted GGM 72B to the Open LLM Leaderboard for benchmarking, I'll send an update post once the results are in!

You can try it out and share your feedback, check out the model page and see what it can do:
👉 rubenroy/Gilgamesh-72B

Would love to hear your thoughts!

posted an update 4 months ago

Post

1950

New sneak preview of my next release! Raiden is a deepseek-ai/DeepSeek-R1 synthetic dataset that uses creative-reasoning and analytic-reasoning prompts!

This preview release has the first 5.8k rows, all responses generated using DeepSeek's 685b parameter R1 model: https://huggingface.co/datasets/sequelbox/Raiden-DSR1-PREVIEW

Enjoy this look at R1's reasoning skills! Full dataset coming soon.

reacted to victor's post with 🚀 4 months ago

Post

3222

Finally, an open-source AI that turns your lyrics into full songs is here—meet YuE! Unlike other tools that only create short clips, YuE can make entire songs (up to 5 minutes) with vocals, melody, and instruments all working together. Letsss go!

m-a-p/YuE-s1-7B-anneal-en-cot

posted an update 5 months ago

Post

2363

A general FYI that Valiant Labs no longer has an X account. This is a business decision. Many other businesses seem to be making the same decision right now.

You can follow my account on Bluesky for updates on Shining Valiant 3, other Valiant Labs models, my open-source datasets, etc: https://bsky.app/profile/sequelbox.bsky.social

back to building :)

posted an update 5 months ago

Post

1401

NEW RELEASE: the sequelbox/Tachibana-QVQ dataset is here! Code-reasoning and code-instruct data generated with Qwen/QVQ-72B-Preview

Come check out QVQ's coding skills!

for everyone to use!

more QVQ and Llama 3.1 405b datasets coming soon :)

reacted to DawnC's post with ❤️ 5 months ago

Post

2334

🌟 PawMatchAI: Making Breed Selection More Intuitive! 🐕
Excited to share the latest update to this AI-powered companion for finding your perfect furry friend! I've made significant architectural improvements to enhance breed recognition accuracy and feature detection.

✨ What's New?
Enhanced breed recognition through advanced morphological feature analysis:
- Implemented a sophisticated feature extraction system that analyzes specific characteristics like body proportions, head features, tail structure, fur texture, and color patterns
- Added an intelligent attention mechanism that dynamically focuses on the most relevant features for each image
- Improved multi-dog detection capabilities through enhanced spatial feature analysis
- Achieved better precision in distinguishing subtle breed characteristics

🎯 Key Features:
Smart breed recognition powered by advanced AI architecture
Visual matching scores with intuitive color indicators
Detailed breed comparisons with interactive tooltips
Lifestyle-based recommendations tailored to your needs

💭 Project Vision
Combining my passion for AI and pets, this project represents another step toward creating meaningful AI applications. Each update aims to make the breed selection process more accessible while improving the underlying technology.

👉 Try it now: DawnC/PawMatchAI

Your likes ❤️ on this space fuel this project's growth!

#AI #MachineLearning #DeepLearning #Pytorch #ComputerVision #TechForLife

2 replies

posted an update 5 months ago

Post

2174

Check out the early preview of the upcoming Tachibana-QVQ dataset: code-reasoning and code-instruct data generated with Qwen/QVQ-72B-Preview

Link here: sequelbox/Tachibana-QVQ-PREVIEW

more to come :)

1 reply

reacted to m-ric's post with 👀 6 months ago

Post

2613

𝐇𝐮𝐠𝐠𝐢𝐧𝐠 𝐅𝐚𝐜𝐞 𝐫𝐞𝐥𝐞𝐚𝐬𝐞𝐬 𝐏𝐢𝐜𝐨𝐭𝐫𝐨𝐧, 𝐚 𝐦𝐢𝐜𝐫𝐨𝐬𝐜𝐨𝐩𝐢𝐜 𝐥𝐢𝐛 𝐭𝐡𝐚𝐭 𝐬𝐨𝐥𝐯𝐞𝐬 𝐋𝐋𝐌 𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝟒𝐃 𝐩𝐚𝐫𝐚𝐥𝐥𝐞𝐥𝐢𝐳𝐚𝐭𝐢𝐨𝐧 🥳

🕰️ Llama-3.1-405B took 39 million GPU-hours to train, i.e. about 4.5 thousand years.

👴🏻 If they had needed all this time, we would have GPU stories from the time of Pharaoh 𓂀: "Alas, Lord of Two Lands, the shipment of counting-stones arriving from Cathay was lost to pirates, this shall delay the building of your computing temple by many moons "

🛠️ But instead, they just parallelized the training on 24k H100s, which made it take just a few months.
This required parallelizing across 4 dimensions: data, tensor, context, pipeline.
And it is infamously hard to do, making for bloated code repos that hold together only by magic.

🤏 𝗕𝘂𝘁 𝗻𝗼𝘄 𝘄𝗲 𝗱𝗼𝗻'𝘁 𝗻𝗲𝗲𝗱 𝗵𝘂𝗴𝗲 𝗿𝗲𝗽𝗼𝘀 𝗮𝗻𝘆𝗺𝗼𝗿𝗲! Instead of building mega-training codes, Hugging Face colleagues cooked in the other direction, towards tiny 4D parallelism libs. A team has built Nanotron, already widely used in industry.
And now a team releases Picotron, a radical approach to code 4D Parallelism in just a few hundred lines of code, a real engineering prowess, making it much easier to understand what's actually happening!

⚡ 𝗜𝘁'𝘀 𝘁𝗶𝗻𝘆, 𝘆𝗲𝘁 𝗽𝗼𝘄𝗲𝗿𝗳𝘂𝗹:
Counting in MFU (Model FLOPs Utilization, how much the model actually uses all the compute potential), this lib reaches ~50% on SmolLM-1.7B model with 8 H100 GPUs, which is really close to what huge libs would reach. (Caution: the team is leading further benchmarks to verify this)

Go take a look 👉 https://github.com/huggingface/picotron/tree/main/picotron

1 reply