This is a harmful slur, it’s a proxy for GPUpoor.

every time I try to generate synthetic data with gpt/claude for alignment research, I spend 30 minutes convincing them I'm not an evil person... bro, i cannot cause an ai apocalypse by finetuning llama-1b, trust me... #gpupoor

My man... ps. any way to speed up the map part than just sequential? #gpupoor #m116gb

#gpupoor #plsdonate

#GPUPOOR #007FirstLightRTX

This baby is crawling 2 billion pages per month and hitting 1,400 tokens/s. The room stays at 30 °C (86°F) with a signature Founder Mode scent of ionized ozone and 'Eau de Silicon'. #WebSearch #LLMs #GPUPoor #Naget

🥳🥳 I love how for this model's param size (229B) the weights are ~150Gb - super good for trying out 50-75% compression using SparseGPT so it can run on 1xH100 -we're trying our best to do this for us GPUPoor folk by going through the fire of CUDA hell: github.com/Pranshu-Bahadur/4…

Keep us posted (crying as a gpupoor)

3D require compute, designer are GPUpoor

. @AskPerplexity can Jensen Huang @NVIDIAGeForce give me and my friend @shine_gupta17 we are GPUpoor 😭😭😭

Big shout out to lium.io for very generous grant, it will speed up my project development. It also allows me to not be a GPUpoor for a pretty good amount of time. Thanks!

About to be slightly less #gpupoor

aku bila kena buat ai research with a single 3090 #gpupoor

Seriously guys the latest days with these enormous models are spamming the "main page" of the model search (especially for new users) Also, "mrdmancher" or however he's called - when you filter "Recently Updated" in search, this guy is a spammer -> we need "model size" bar and also "exclude from search" (to for example write in the search bar that we look for "LLAMA" but also we don't want "mrdmancher" string to be included in the results) You know, like with grep and the -e flag Seriously, even for me who uses your app from long time, it's hard to find useful latest model these days... BTW Hugging Face recently added a "model max size" bar in search to filter results for GPUpoor people with just 6GB VRAM it will be fast to implement and I believe the user experience (ESPECIALLY FRESH USERS!) will gain a lot of simplicity, instead of current confusion that exists Wish you the best and greetings from Poland 💪

Tried using actual weights of Gemini 3b pt 224 on Collab and kaggle but ended up wasting 4 hours So now will only use the random weights , images and text just for making sure the architecture works in future I am gpupoor 😭😭😭

highly relatable as a macbook air user 🥲🥲 i am gpupoor while being a gpupaglu

Polaris aims to eliminate the category of #GPUPOOR - Labs, teams and researchers across the globe, developed and so called “underdeveloped” countries, this is for you! - need to run a quick experiment? Need some machines? Grab a pod on Polaris. It’s super easy, beautiful, 100% free & incentivised by the most powerful intelligence network on the planet #bittensor. @TAOCommunityHub @mogmachine

In the process of generating complete frame-by-frame Catcher glove data for April 2024 (over 120,000 videos need processed). If you conduct public baseball research and would like early access (free ofc) or would like to donate compute to generate more data, reach out! #GPUPoor

LoRI is an evolution of LoRA that: •Uses random projections & sparse masks •Cuts (LoRA) trainable params by up to 95% •Reduces task interference •Enables continual & multi-task learning •Outperforms/matches LoRA & full fine-tuning #gpupoor

Another Tuesday morning. Another "Due to unexpected capacity constraints, Claude is unable to respond to your message. Please try again soon." #GPUpoor