HackerTrans
TopNewTrendsCommentsPastAskShowJobs

110

no profile record

Submissions

MiniMax M2.1: Built for Real-World Complex Tasks, Multi-Language Programming

minimaxi.com
228 points·by 110·7개월 전·81 comments

Show HN: Khoj – Chat offline with your second brain using Llama 2

github.com
565 points·by 110·3년 전·150 comments

comments

110
·2년 전·discuss
Potential solutions:

1. AI video watermarks that carry over even if a video of the AI video is taken

2. Cameras that can see AI video watermarks and put an AI video watermark on the videos of any AI videos they take
110
·2년 전·discuss
Umm, customary, tongue-in-cheek reference to McCarthy's proposal for a 10 person research team to solve AI in 2 months (over the Summers)[1]. This was ~70 years ago :)

Not saying we're in necessarily the same situation. But it remains difficult to evaluate effort required for actual progress.

[1]: https://www-formal.stanford.edu/jmc/history/dartmouth/dartmo...
110
·3년 전·discuss
Yes, this seems to be a common concern. We're trying to see how best to crisply address it. But yes, Khoj can be used even with your internet turned off
110
·3년 전·discuss
We're still trying to figure out the right balance between configurability and ease of use/maintenance.

The 7B version was a decent enough starting point in terms of what it can answer (and way fewer folks can run a 13B on their machine).

If you really want you can just replace the 7B model file with the 13B one under the ~/.cache/gpt4all directory on your device and it should just work.
110
·3년 전·discuss
Yes, of course
110
·3년 전·discuss
It should work but not sure how fast. Best way to find out would be to try it.

We also have some fixes and perf improvements we plan to release later today in version 0.10.1
110
·3년 전·discuss
Thanks for the feedback. Does your machine have a GPU? 32GB CPU RAM should be enough but GPU speeds up response time.

We have fixes for the seg fault[1] and improvement to the query speed[2] that should be released by end of day today[3].

Update khoj to version 0.10.1 with pip install --upgrade khoj-assistant later today to see if that improves your experience.

The number of documents/pages/entries doesn't scale memory utilization as quickly and doesn't affect the search, chat response time as much

[1]: The seg fault would occur when folks sent multiple chat queries at the same time. A lock and some UX improvements fixed that

[2]: The query time improvements are done by increasing batch size, to trade-off increased memory utilization for more speed

[3]: The relevant pull request for reference: https://github.com/khoj-ai/khoj/pull/393
110
·3년 전·discuss
Khoj provides exactly that; it runs on your machine, none of your data leaves your machine and it has a web interface to chat
110
·3년 전·discuss
Firstly, my previous statement needs quite a lot of clarification:

Khoj works more like an incremental, natural language version of org-agenda-search (or projectile) rather than isearch.

You've to configure which files it should index first. You can then use natural language to search those files with a search-as-you-type experience.

This is not the same as isearch that just searches the current file for keyword matches.

Second, khoj doesn't index source code and the default search models don't work well with code files.

But yeah someone should implement a natural language isearch as a standalone tool (as suggested by parent comment). It'd be super-useful
110
·3년 전·discuss
Khoj exposes a local API. Hopefully that makes it easy to integrate with Raycast, Alfred (or even Spotlight)?
110
·3년 전·discuss
Yeah, this would be ideal for Mac users. Just need to look into what is required and how much work it is
110
·3년 전·discuss
Thanks for the feedback! Someone else mentioned this issue the other day as well. I'll fix this issue on the landing page soon
110
·3년 전·discuss
To use Chat with non-english sources you'll need to enable OpenAI. Offline chat with Llama 2 can't do that yet.

And Search can be configured to work with 50+ languages.

You'll just need to configure the asymmetric search model khoj uses to paraphrase-multilingual-MiniLM-L12-v2 in your ~/.khoj/khoj.yml config file

For setup details see http://docs.khoj.dev/#/advanced?id=search-across-different-l...
110
·3년 전·discuss
Thanks! :)
110
·3년 전·discuss
Buggy for setup? We've done some improvements and have desktop apps (in beta) too now to simplify this. Feel free to report any issues on the khoj github. I can have a look.

Yes, we don't do optimizations on the query encoding yet. So SBERT just re-encodes the whole query every time. It gets results in <100ms which is good enough for incremental search.

I did create a plugin system, so that a data plugin just has to convert the source data into a standardized intermeditate jsonl format. But this hasn't been documented or extensively tested yet.
110
·3년 전·discuss
Khoj chat with Llama2 will not work with non-english languages though. You'll have to enable OpenAI for that
110
·3년 전·discuss
Search should work with Slavic languages including Russian and 50+ other languages.

You'll just need to configure the asymmetric search model khoj uses to paraphrase-multilingual-MiniLM-L12-v2 in your ~/.khoj/khoj.yml config file

See http://docs.khoj.dev/#/advanced?id=search-across-different-l...
110
·3년 전·discuss
You do not need to break the bank to use Khoj for local chat, 16Gb RAM should be good enough
110
·3년 전·discuss
Oh yeah, I used to use Google Desktop too. It was awesome for it's time
110
·3년 전·discuss
Khoj uses Llama 2 7B, 4bit quantized. So it just needs 3.5Gb of RAM (GPU or System) [1].

Khoj and your other apps need more RAM themselves, so practically 8GB of System or GPU RAM should suffice.

Khoj has been tested with CUDA and Metal capable GPUs. So Nvidia and Mac M1+ GPUs should work. I'm think it'll work with AMD GPUs out of the box too but let me know if it doesn't for you? I can look into what needs to be done to get that to work.

[1]: The calculation is [params] * [bytes] GB RAM, so 7 * 0.5 = 3.5Gb