Top all-time · #8801
Featherless AI dev blog
Company blog · substack.recursal.aiai
Featherless AI dev blog is the company blog at substack.recursal.ai, writing about ai. Since 2024, 9 different Hacker News readers have linked it 11 times, most often “Attention is NOT all you need: Qwerky-72B trained using only 8 AMD MI300X GPUs”. It is still writing: the last post was on 2025-05-23, and you can follow it by feed.
Too few sampled comments (4) to say how often HN recommends it.?
Why HN links it
HN readers link Recursal's posts on its RWKV and EagleX language model releases, often as the source for model size, training data, and benchmark claims. The comments praise none of the writing.
Summarised by Claude from a sample of 4 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
A blog for big updates from the product, engineering and research teams at Featherless AI.
Most-linked posts?
- Attention is NOT all you need: Qwerky-72B trained using only 8 AMD MI300X GPUs5 people · 5 mentions · first linked 2025-03-24
- EagleX 1.7T: Soaring past LLaMA 7B 2T in both English and Multi-lang evals3 people · 3 mentions · first linked 2024-03-16
What Hacker News says?
https://substack.recursal.ai/p/eaglex-17t-soaring-past-llama... this one claims to have been trained only on permissively licensed data.
Comment linking /p/eaglex-17t-soaring-past-llama-7b
on “Grok”Submitted EagleX 1.7T: Soaring past LLaMA 7B 2T in both English and Multi-lang evals as a story · 36 points
HN discussionI have no idea how it works actually (in google) but I wouldn't be surprised if it was just post-training because recently RWKV people did something similar: They replaced the whole attention mechanism with WKV (forward-only linear attention), and created such franken-stein just by post-training. The big wow moment about that is that it sort of implies that most of the useful knowledge is in the FFN, and attention i…
Comment linking /p/qwerky-72b-and-32b-training-large
on “Gemini Diffusion”Submitted Attention is NOT all you need: Qwerky-72B trained using only 8 AMD MI300X GPUs as a story · 20 points
HN discussionCurrently the strongest RWKV model is 32B in size: https://substack.recursal.ai/p/q-rwkv-6-32b-instruct-preview This is a full drop in replacement for any transformer model use cases on model sizes 32B and under, as it has equal performance to existing open 32B models in most benchmarks We are in works on a 70B, which will be a full drop in replacement for most text use cases
Comment linking /p/q-rwkv-6-32b-instruct-preview
on “RWKV Language Model”Submitted Qwerky 72B – A 72B LLM without transformer attention as a story · 4 points
HN discussionblog post: https://substack.recursal.ai/p/eaglex-17t-soaring-past-llama...
Comment linking /p/eaglex-17t-soaring-past-llama-7b
on “EagleX 7B with 1.7T tokens beats llama2 7B (2T tokens) on English evals”Submitted Qwerky: Attention is not what you need? RWKV mashed into QwQ models as a story · 3 points
HN discussionSubmitted The most reliable AI agent that works – where Claude, Gemini, and o3 fail as a story · 1 points
HN discussionSubmitted Training large attention free models as a story · 1 points
HN discussionSubmitted Please stop throwing money at AI founders with no commercial plan, besides AGI as a story · 1 points
HN discussion
