furiousteabag

7 ms·

1.
by furiousteabag 1y ago
same, did you figure it out?
2.
by furiousteabag 2y ago
Another thing we are trying to understand is whether the 2D element adds value to the simulation. A simpler option would be a pure text/chat interface. Still, the hypothesis here is that it is easier to comprehend what's going on
3.
by furiousteabag 2y ago
Hey hey! We tried Clay v1 with 768 embeddings size using your tutorials. We then split NAIP SF to chips and indexed them. Afterwards, we performed image-to-image similarity search like in your explorer. We tried to search for bridges, beach
4.
by furiousteabag 2y ago
Curious to see what other things you will simulate in the future! Shameless plug: recently we've built a demo that allows you to search for objects in San Francisco using natural language. You can look for things like Tesla cars, dry p
5.
by furiousteabag 2y ago
Thanks for sharing Brooklyn text demo. Haven't seen it! Captioning images using VLM would definitely help as an additional conditional feature. Maybe it even would be enough to use only embeddings of captions to do search! We chose aer
6.
Show HN: Search San Francisco using natural language(demo.bluesight.ai)
12 points by furiousteabag 2y ago|2 comments
7.
Show HN: Search San Francisco satellite imagery using natural language(demo.bluesight.ai)
5 points by furiousteabag 2y ago|0 comments
8.
Hawking's Time Traveller Party(en.wikipedia.org)
2 points by furiousteabag 2y ago|0 comments
9.
Is Entrepreneur First worth the time? Honest review of the acceleration program(asmirnov.xyz)
2 points by furiousteabag 2y ago|0 comments
10.
Startup accelerators accepting applications now and terms(twitter.com)
2 points by furiousteabag 2y ago|0 comments
11.
GitHub Commit Annotation Bug(twitter.com)
1 points by furiousteabag 2y ago|0 comments
12.
by furiousteabag 3y ago
I really like using pandoc as a build system [1] for my personal website to convert .md to .html. I can use templates, automatically generate a table of content and run some lua scripts to get the desired result, such as clickable headers.
13.
by furiousteabag 3y ago
You are correct, training sorely in fp16/bf16 can lead to imprecise weight updates or even gradients turning to zero. Because of that, mixed precision is used. In mixed precision training, we keep a copy of the weights in fp32 (master
14.
by furiousteabag 3y ago
Mixed precision is a default method to pretrain and full fine tune right now. It is especially good in transformers, because they have memory bottleneck in activations (outputs of intermediate layers stored for backprop), and running forwar
15.
by furiousteabag 3y ago
There is no option to select quantized version yet. Will work on that!
16.
by furiousteabag 3y ago
By default, SGD w momentum is enabled as optimizer. You may try selecting Adam and it will list second moments as well.
17.
Show HN: I made a GPU VRAM calculator for transformer-based models(vram.asmirnov.xyz)
135 points by furiousteabag 3y ago|38 comments
18.
Lessons from failing my AI startup(asmirnov.xyz)
1 points by furiousteabag 3y ago|0 comments
19.
Ask HN: Clarify VRAM usage during LLM forward pass
1 points by furiousteabag 3y ago|0 comments
20.
by furiousteabag 3y ago
It doesn't really matter if the app claims to use E2E when it actually discloses message content [0] [1]. WhatsApp is also filled with backdoors [2]. [0] https://therecord.media/fbi-document-shows-what-data-can-be-...
21.
by furiousteabag 3y ago
This may sound stupid, but from my perspective renting random VMs on vast.ai is safe in general and might be safer than using traditional cloud providers in particular. Consider this: on your VM a new image starts several times a day, each
22.
by furiousteabag 3y ago
In IM, there's a balance between total privacy and widespread use. Apps like Signal offer high privacy but have fewer users, while popular ones like WhatsApp are less secure. Telegram lies somewhere in between, offering a level of priv
23.
by furiousteabag 3y ago
It's true that fine-tuning models on personal messages could be simplified, but many, like myself, can't use third-party services due to sensitive data in our messages. I'm curious if others face this trust issue and how it m
24.
by furiousteabag 3y ago
I agree that usually 'more is more' for training LLMs. However, for fine-tuning with limited data, it seems crucial to focus the task as much as possible. Since the model still encounters these masked sentences in the data, it eff
25.
by furiousteabag 3y ago
First I download the weights of the base pre-trained model to the VM instance. Then I upload my data there. Afterward, I fine-tune either LoRA or full and when training finishes, from the VM instance I download the adapters in case of LoRA
26.
by furiousteabag 3y ago
I think incorporating knowledge from other apps is a good next step because the model definitely lacks the context of what is going on right now. The nature of instant messaging is that most of the messages are about what is happening right
27.
by furiousteabag 3y ago
I've been using vast.ai for a very long time. It is like a GPU marketplace, where people rent and lease GPUs. There are a lot of VMs with 4090, and beasts like 8xA100 80GB are also available from time to time.
28.
Learnings from fine-tuning LLM on my Telegram messages(asmirnov.xyz)
216 points by furiousteabag 3y ago|69 comments
29.
Failing AI Startup in 2023(asmirnov.xyz)
7 points by furiousteabag 3y ago|0 comments
30.
Show HN: I made a CLI link manager with Git storage(github.com)
1 points by furiousteabag 4y ago|0 comments

More ›