tarruda

28 ms·

391.
by tarruda 3y ago
Seems like he's completely emotion driven at this point. I doubt anyone advising rationally would agree with sending this tweet
392.
by tarruda 3y ago
Microsoft under Nadella always wins
393.
by tarruda 3y ago
This tweet achieves absolutely nothing except give the impression of a weak leadership and that firing Sam Altman was done on a whim.
394.
by tarruda 3y ago
November 17th OpenAI blog post: "The board no longer has confidence in his ability to continue leading OpenAI." The fact that they agreed to join as MS employees kinda proves that money was a big motivator.
395.
by tarruda 3y ago
Did he say GPT-4 API costs OpenAI $3/token?
396.
by tarruda 3y ago
I think it is pretty simple to decompile Lua bytecode
397.
by tarruda 3y ago
I'm not familiar with Common Lisp or its systems programming features, but they can access any C API through LuaJIT FFI: https://luajit.org/ext_ffi.html LuaJIT is so good that it is actually able to inline calls to C o
398.
by tarruda 3y ago
I'm curious about this web page: Did you compile jq targeting WASI/WASM to run an in-browser version?
399.
Thompson's Satan rules our computers(pluralistic.net)
1 points by tarruda 3y ago|0 comments
400.
by tarruda 3y ago
Recommended reading order by Mike Pall (author of LuaJIT): https://old.reddit.com/r/programming/comments/63hth/ask_redd...
401.
Reading Lua 5.1 source code(old.reddit.com)
3 points by tarruda 3y ago|1 comments
402.
by tarruda 3y ago
Also not certain about details of how it works, but there are open source youtube frontends that manage to do this, while also adblocking: - https://github.com/TeamPiped/Piped - https://newpipe.net/ II
403.
The Techno-Optimist Manifesto(andrewkelley.me)
1 points by tarruda 3y ago|2 comments
404.
by tarruda 3y ago
Yes, it is a instruction tune dataset: https://huggingface.co/datasets/Open-Orca/OpenOrca It felt different from the official Mistral7B-Instruct. One of the highlights with the OpenOrca version is that you can ste
405.
by tarruda 3y ago
> K3S is for orchestration across nodes, and by design you can't run services on the master for security reasons unless you manually change it. That has not been my experience with k3s. Unlike normal distributions, k3s servers do al
406.
by tarruda 3y ago
It is better than llama 2 7b and 13b. I tried the OpenOrca fine tune and it is very good, even when 4-bit quantized
407.
by tarruda 3y ago
8tk/s on 34b? I've managed to run Codellama instruct 13b with my laptop's RTX 3070 (8gb VRAM) at 6tk/s by offloading 27 layers into the GPU with llama.cpp I've been considering getting a macbook for running 34b+ LLM
408.
by tarruda 3y ago
Are you using apple silicon? How much RAM do you have, and how many tokens/second with codellama 13b?
409.
by tarruda 3y ago
If the rumors around GPT4 being a mixture of expert models are true, the this comparison is not fair. What would be interesting is compare GPT4 at a certain task with a small model fine tuned for that task.
410.
by tarruda 3y ago
> Or is there a way to get better results by fine tuning? It seems that the general wisdom around LLMs is that you can get very good performance on small models if you fine tune for a specific task. In the case of code generation, I thin
411.
by tarruda 3y ago
How does one decide which of these modules shouls be fine tuned for different tasks?
412.
by tarruda 3y ago
Very cool, but I really like wall-mounting these ARM SBC near my router. This one seems unfit for that purpose. The argon one raspberry pi 4 case is very good for wall mounting, though it connects with the m.2 drive via USB 3, which kinda s
413.
by tarruda 3y ago
I haven't, but didn't write the article. I don't understand what these modules mean for fine-tuning Here someone does Alpaca LoRA fine-tuning on Llama 7b using only q/v modules: https://colab.research.google.
414.
by tarruda 3y ago
For chat, I suggest trying the OpenOrca finetune, the performance is amazing and it responds very well to system prompt: https://huggingface.co/Open-Orca/Mistral-7B-OpenOrca It runs very well on my laptop's RTX 30
415.
by tarruda 3y ago
I'd say the original Alpaca paper is a good source of inspiration on how to create datasets, they even shared the script used to generate data using OpenAI API. One thing that came to mind for creating datasets for other programming la
416.
by tarruda 3y ago
> AI can truly take over programming jobs (notice I did not say Software Engineering on purpose) Can you elaborate on the difference?
417.
Fine-Tuning Mistral7B on Python Code(wandb.ai)
61 points by tarruda 3y ago|11 comments
418.
by tarruda 3y ago
This article from 2020 says Debian doesn't need money: https://www.theregister.com/2020/09/10/debian_project_addres...
419.
by tarruda 3y ago
Not scoffing, performance is nice to have. I really like how Bun replaces all development tools, is much faster for package management and bundling. The fast startup is also a bonus and means it could be used for writing command line tools
420.
by tarruda 3y ago
Yes. I don't know where this obsession over making JS fast comes from, considering most web applications are http glue code to a database server and spend most of time waiting on I/O Biggest selling point of Bun is the potential i

More ›