azeirah

13 ms·

181.
by azeirah 3y ago
From my experience when using llama.cpp, using min-p sampling gives far better results than the defaults, for example (note this is using deepseek-coder 33b) Prompt: "Write the fibonacci function in Python3" min_p 0.05 def f
182.
by azeirah 3y ago
ComfyUI works perfectly fine with ROCm on Linux. Using it with this krita plugin also works flawlessly. The docs are simply incorrect in saying that it's Windows-only. I assume it's the case because the automatic ComfyUI installer
183.
by azeirah 3y ago
Does anyone know if there are any portable bluetooth keyboards (think usable in the train) with a trackpoint? The only ones I can find have a trackpad, but I think a trackpoint would actually be a lot nicer.
184.
by azeirah 3y ago
Yeah this becomes very noticeable when you're deliberately learning facts with Anki. For example when trying to learn immunology, "This bacterium causes neural bacteriumeritis", score 100% on the set of cards after a few sess
185.
by azeirah 3y ago
You could also just buy a very large amount of their coral consumer TPUs :D
186.
by azeirah 3y ago
I think it's still a developing technique. I'm not following RAG super closely, but I keep myself somewhat updated through Sam Witteveen's videos on YouTube: https://www.youtube.com/@samwitteveenai His channe
187.
by azeirah 3y ago
You shouldn't need to fine-tune it at all. Whisper supports adding prompts -- not to be confused with GPT-style prompts -- these prompts let you specify the "style" of output the model should give. So if you're giving in
188.
by azeirah 3y ago
Given that I keep getting ads on reddit about some API to predict age and gender from a given dataset of users and the OP of that ad keeps saying that it's hugely succesful, this business is practically guaranteed to succeed. Seriously
189.
by azeirah 3y ago
For what it's worth, I do feel like text editing is a big problem on Mobile. This is why I got myself a samsung note-line phone, the stylus greatly improves the text editing experience by basically being about as precise as a cursor an
190.
Medusa: Framework for Accelerating LLM Generation with Multiple Decoding Heads(sites.google.com)
2 points by azeirah 3y ago|1 comments
191.
by azeirah 3y ago
I wrote some Imba code for fun a while ago. Nothing special, just an HTML clock or something. It was quite nice to work with, really simple. Though I was just not convinced by its ecosystem, who's gonna use this?
192.
by azeirah 3y ago
I don't disagree with your proof but I would like to add that your example doesn't mean it's not practical to recursively summarise a couple times. One bit of information can mean more the more parameters an LLM has. The text
193.
by azeirah 3y ago
Love it! Learned a lot already in the first few articles.
194.
by azeirah 3y ago
Huh? This is not an alternative devtools application, it's literally just a large collection of tutorials and tips on how to get the most out of your browser's devtools.
195.
by azeirah 3y ago
Microsoft is a hosting partner, there's an Azure service for hosted private LLaMa inference for business. Being a go-to hosting provider for SoTA AI is of course a very good thing for Microsoft.
196.
by azeirah 3y ago
I know plenty of open-source projects who list and thank every individual contributor. The website could do that too!
197.
by azeirah 3y ago
What do herbicides have to do with this study? They added gluten to the diet, not wheat.
198.
by azeirah 3y ago
I do stuff like this sometimes when I have some csv or something and need it in JSON. Could easily be done with 1 line of bash or js or python or whatever but... it's easier to just let the LLM do it for me :|
199.
by azeirah 3y ago
Another big difference is Google's name and marketing behind it. It had its own website.
200.
by azeirah 3y ago
What do you mean by wizard+vicuna? I use wizardlm myself right now and have used vicuna in the past. Do you mean there's a model trained on the datasets of both wizardlm and vicuna?
201.
by azeirah 3y ago
Haha, I wasn't worried about the GPU price. I was worried about the price of 8 decent quality monitors. I have a decent salary but I don't have throwaway money to just spend 2k on monitors. Or on anything for that matter, that req
202.
by azeirah 3y ago
Note that despite this being a python-specific slide deck, generators and iterators are also present in many other languages, including but not limited to Rust and JS. The concepts matter more than the chosen language in this deck. I learne
203.
by azeirah 3y ago
8 x 4K monitors, I'm so jealous wow. I wish my work would give me even one, I'm developing on 2 full hd monitors :( I even do some design work so it's honestly pretty bad.
204.
by azeirah 3y ago
Have you tried the new profiler since you're talking about performance analysis? It's insanely in-depth.
205.
by azeirah 3y ago
Firefox dev tools have been letting me down the past few months. I'm working on a pretty nasty (legacy and poorly optimised) but otherwise still rather "normal" website at work. Opening the dev tools makes Firefox hang for al
206.
by azeirah 3y ago
https://github.com/ggerganov/llama.cpp/issues/2262 Likely needs to be updated Edit: Only the case for the 34B and 70B models. 7B and 13B run as-is. You can download the GGML model already https://h
207.
by azeirah 3y ago
Thank you for the doorknob example. It's perfectly on-point.
208.
by azeirah 3y ago
Don't know any stats from the top of my head, but I recall that oesophageal cancer is caused by drinking _hot_ tea, not by drinking tea per sé.
209.
by azeirah 3y ago
I would appreciate one! mail at laura.fm
210.
by azeirah 3y ago
Yeah for sure. I do believe AI has a lot of potential to break through various hard problems we currently face in science and medicine. I just really dislike the idea of the "omniscient super-intelligent ruler of chaos" vision of

More ›