@reach_vb

–

08 December 2024 20h28

Good call – I quite liked OuteTTS recently, thinking the LLM direction is still a bit slept on.

Llama Stack Building Open Ecosystem and Business Opportunities

By

–

08 December 2024 20h27

Quite bullish on the whole llama stack – Even if 50% of this wishlist is completed we'd be raising a huge array of open ecosystem, business around it.

Pre-training Data Diversity Drives Major AI Model Improvements

By

–

08 December 2024 20h26

I think a lot depends on how much pre-training data went into and how diverse was it – we made a massive jump from the models in 2023 to 2024. I'm quite sure the jump from 2024 to 2025 will be quite significant too!

Pre-training Dataset Distinguishes from LLM Models

By

–

08 December 2024 11h04

It’s a pre-training dataset – not a LLM.

Hugging Face Releases Fineweb-2 Dataset for AI Training

By

–

08 December 2024 10h07

Check it out here: https://
huggingface.co/datasets/Huggi
ngFaceFW/fineweb-2
…

FineWeb 2.0: 3 Trillion Token Multilingual Training Corpus Released

By

–

08 December 2024 10h07

FineWeb 2.0 – 8 Terabytes, 3 Trillion tokens, 1000 languages – simply the best multilingual pre-training corpus out there! Available under a commercially permissive license!

Meta’s 2GW Data Center Fuels Next-Gen Llama AI Models

By

–

07 December 2024 23h52

Meta is cooking! 2 GW+ data centre is quite crazy! In case anyone from meta is listening, here’s my wish list: 1. Full duplex Llama – I’m talking Moshi style speech to speech Llama, w/ voice control
2. Smaller Llama – Sub 1B, multilingual, large context
3. Text to Speech

7 December 2024

Major VLM Revolution: Open-Source Models from Google, OpenGVLabs, Qwen, Microsoft

By

–

07 December 2024 11h07

VLMs are going through quite an open revolution AND on-device friendly sizes: > Google DeepMind w/ PaliGemma2 – 3B, 10B & 28B > OpenGVLabs w/ InternVL 2.5 – 1B, 2B, 4B, 8B, 26B, 38B & 78B > Qwen w/ Qwen 2 VL – 2B, 7B & 72B > Microsoft w/ FlorenceVL – 3B & 8B (Links below)