Hacker News: minxomat

New comment by minxomat in "Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA"

minxomat — Wed, 22 Mar 2023 09:43:31 +0000

Is this related in any way to the actual ChatLLama project that is developed here: https://github.com/nebuly-ai/nebullvm ?

New comment by minxomat in "TextSynth Server"

minxomat — Fri, 17 Mar 2023 08:02:10 +0000

The former. Makes sense for their business model.

New comment by minxomat in "Duolingo Max, a learning experience powered by GPT-4"

minxomat — Wed, 15 Mar 2023 09:59:09 +0000

Follow up recommendation: for Spanish, “Madrigals Key” is language transfer in book form. It can in fact be used as a workbook for this course.

New comment by minxomat in "From Books to Knowledge Graphs"

minxomat — Tue, 14 Mar 2023 22:18:58 +0000

Sure. Sent a ping.

New comment by minxomat in "The Bing AI bot has been secretly running GPT-4"

minxomat — Tue, 14 Mar 2023 20:35:48 +0000

We're not far off: https://imgur.com/a/nJxxcUg

New comment by minxomat in "Khan Academy integrates GPT-4 as every student’s customized tutor"

minxomat — Tue, 14 Mar 2023 19:07:03 +0000

A plot point in an episode of Person of Interest, where AGI "Samaritan" funds a charity (free tablets for students, Samaritan access pre-installed) to take over education and recruit more mercenaries.

New comment by minxomat in "From Books to Knowledge Graphs"

minxomat — Tue, 14 Mar 2023 16:46:54 +0000

I'm building this, for (mostly) non-scientific non-fiction works (books, articles, news, etc.). Launching soon, with about 7,500 books indexed.

Generally, what I found useful to build a graph between "topics" or entities was to use a HyDE[1] prompt to generate possible distinct definitions and then build a nearest-neighbor network from that. This successfully identifies related concepts in the truly abstract sense rather than literal entities.

[1] https://summarity.com/hyde

New comment by minxomat in "ChatGPT's API is so good and cheap, it makes most text generating AI obsolete"

minxomat — Sun, 12 Mar 2023 00:22:38 +0000

The whole point of the LLaMa paper is that large models are undertrained and oversized.

New comment by minxomat in "Running LLaMA 7B on a 64GB M2 MacBook Pro with Llama.cpp"

minxomat — Sat, 11 Mar 2023 15:08:03 +0000

A laptop and a desktop (Mac Studio)

Hypothetical Embeddings Explained

minxomat — Sat, 11 Mar 2023 13:42:39 +0000

Article URL: https://summarity.com/hyde

Comments URL: https://news.ycombinator.com/item?id=35108191

Points: 1

# Comments: 0

New comment by minxomat in "Running LLaMA 7B on a 64GB M2 MacBook Pro with Llama.cpp"

minxomat — Sat, 11 Mar 2023 12:45:44 +0000

This has been fixed almost 2 days ago now. It’s literally mentioned at the top of the repo.

New comment by minxomat in "Llama.cpp: Port of Facebook's LLaMA model in C/C++, with Apple Silicon support"

minxomat — Sat, 11 Mar 2023 12:34:57 +0000

The full use case includes quantisation, which the repo points out uses a large amount of system RAM. Of course that’s not required if you skip that step.

New comment by minxomat in "Llama.cpp: Port of Facebook's LLaMA model in C/C++, with Apple Silicon support"

minxomat — Sat, 11 Mar 2023 12:33:13 +0000

You’re missing something. Both SHP (https://huggingface.co/datasets/stanfordnlp/SHP) and OpenAssistant datasets are referenced.

And the TOS violation might be the case, the project nevertheless has a mode to use OpenAI in the fine tuning steps.

New comment by minxomat in "Running LLaMA 7B on a 64GB M2 MacBook Pro with Llama.cpp"

minxomat — Sat, 11 Mar 2023 12:28:29 +0000

With 16 threads, about 140ms per token for 30B, 300ms per token for 65B

I should also mention that 65B should be able to run on 64GB systems. Total system memory consumption on M1 Ultra is about 67GB when running nothing else.

New comment by minxomat in "Llama.cpp: Port of Facebook's LLaMA model in C/C++, with Apple Silicon support"

minxomat — Sat, 11 Mar 2023 09:49:37 +0000

So if I'm reading this right, 65B at 4bit would consume around 20GB of VRAM and ~130GB of system RAM?

New comment by minxomat in "Llama.cpp: Port of Facebook's LLaMA model in C/C++, with Apple Silicon support"

minxomat — Sat, 11 Mar 2023 09:38:03 +0000

There are open datasets (see the chatllama harness project and its references). You can of course also cross train it using actual ChatGPT.

New comment by minxomat in "Why Write?"

minxomat — Sat, 11 Mar 2023 08:38:50 +0000

AdGuard

New comment by minxomat in "Show HN: Llama-dl – high-speed download of LLaMA, Facebook's 65B GPT model"

minxomat — Sun, 05 Mar 2023 17:03:43 +0000

> not really competitive with ChatGPT

That's impossible to judge. LLama is a foundational model. It has received neither instructional fine tuning (davinci-3) nor RLHF (ChatGPT). It cannot be compared to these finetuned models without, well, finetuning.

New comment by minxomat in "Optical Transformers"

minxomat — Thu, 23 Feb 2023 12:49:38 +0000

GPT-3 might be the ultimate contender for https://en.m.wikipedia.org/wiki/Hutter_Prize - now if we could only find the prompts to recover it all

New comment by minxomat in "Vikings went to Mediterranean for ‘summer jobs’ as mercenaries, left graffiti"

minxomat — Thu, 23 Feb 2023 12:28:26 +0000

Everything old is new again … https://en.m.wiktionary.org/wiki/juvenoia