HackerTrans
トップ新着トレンドコメント過去質問紹介求人

immortal3

no profile record

投稿

Improving LLM Inference with Continuous Batching: Orca Through Tinyorca

junupark.xyz
2 ポイント·投稿者 immortal3·3 か月前·0 コメント

[untitled]

1 ポイント·投稿者 immortal3·4 か月前·0 コメント

Bits-per-Byte (BPB): a tokenizer-agnostic way to measure LLMs

dipkumar.dev
1 ポイント·投稿者 immortal3·9 か月前·0 コメント

Creativity Is a Luxury

dipkumar.dev
2 ポイント·投稿者 immortal3·11 か月前·0 コメント

GPT-5 Router – Inevitable Future of Chat Interfaces

dipkumar.dev
3 ポイント·投稿者 immortal3·11 か月前·0 コメント

Instruction Aware Embeddings – Why Your Retriever Is Failing

dipkumar.dev
1 ポイント·投稿者 immortal3·昨年·0 コメント

Improving Retrieval in RAG (Via Recall, Precision, and NDCG)

dipkumar.dev
2 ポイント·投稿者 immortal3·昨年·0 コメント

Show HN:AceVocab - Learn and master the vocabulary featured in the GRE/GMAT

acevocab.com
3 ポイント·投稿者 immortal3·2 年前·0 コメント

AWS BedRock – Converse API – A single endpoint for all models?

dipkumar.dev
2 ポイント·投稿者 immortal3·2 年前·0 コメント

Essential Database Design: Five Fields Every Table Must Have

dipkumar.dev
3 ポイント·投稿者 immortal3·2 年前·1 コメント

India issues notice to Google for blocking count over nude childhood photo

deccanherald.com
3 ポイント·投稿者 immortal3·2 年前·0 コメント

Hugging Face raises $235M from investors including Salesforce and Nvidia

techcrunch.com
378 ポイント·投稿者 immortal3·3 年前·203 コメント

Speeding up the GPT with KV cache (memoization)

immortal3.github.io
2 ポイント·投稿者 immortal3·3 年前·0 コメント

コメント

immortal3
·9 か月前·議論
There's another angle to this comparison. Groq and Cerebras use custom chips, but I'm not sure about Together. In this case, Together is sharing results based on the B200 GPU. Another important point is the accuracy of these speed-ups compared to the baseline model. It's known that such tricks reduce accuracy, but by how much? Kimi has already benchmarked several providers. https://x.com/Kimi_Moonshot/status/1976926483319763130
immortal3
·2 年前·議論
Is it possible that US or any other government has participated in the round which might be the reason?
immortal3
·2 年前·議論
Honestly, it doesn't matter for the end user if there are more tokens generated between the AI reply and human message. This is like getting rid of AI wrappers for specific tasks. If the jump in accuracy is actual, then for all practical purposes, we have a sufficiently capable AI which has the potential to boost productivity at the largest scale in human history.
immortal3
·2 年前·議論
After the acquisition, I think Wiz would have to only focus on Google Cloud which might be a major limiting factor in the company's future. But other than that, It surprises me that, a $23B offer is rejected from the perspective of Employees. IPO won't provide the same level of liquidity opportunities.

I have used Octa and it's a decent platform, not a magical one. Creating a similar platform for Google Cloud should be feasible with the level of Google resources.
immortal3
·3 年前·議論
To add further, i do not know what is end goal of hugging face. 1. They have inference API but all cloud provider can implement those in next year. 2. They offer subscription but market-size of subscription is questionable. 3. I hope this new set of funding don't bring problem to them because making money with open source is hard and at this scale of funding it might be even harder. It see what happens in next set of years.