My tablet has an option which switches it into the black and white picture mode. I would say that makes less attractive to get into YouTube or doomscrollread messages in this social network. Also, closer to night time less blue screen light comes in to the eyes.
@akshay_pachaar >𝘀𝗲𝘀𝘀𝗶𝗼𝗻𝘀/ stores per-platform session metadata. 𝘀𝘁𝗮𝘁𝗲.𝗱𝗯 is the SQLite database
Hello Akshay, suppose you are working in two parallel branches and save both sessions into this file, how do you solve merge conflict later?
@HowToAI_@grok could you please read that article carefully and make a skill which helps agents to MITIGATE MODE COLLAPSE AND UNLOCK LLM DIVERSITY when they asked to be creative?
@Ric_RTP LLM is omnidirectional force, you need to build an engine around it so it can do useful work. As clear you see the goal, as precise it delivers you to it.
@elonmusk Netherlands roads quality is very high though there are a number of difficult crossroads in Amsterdam. Ams has speed limit 30 km/h in a big part of the center)
@Suryanshti777 It is true, LLM is like fire, it is omnidirectional. It is required to build the engine around it in order to receive the awaited result
@karpathy Cool, good luck in fighting with QK-norm dominance! The project is interesting itself, are you going to publish details such as nanochat + softcap mod, agent prompts, orchestration and API integration details?
@ozgurdeniz358@akshay_pachaar Estimatelly miinimum 101 GB for 3-bit, 96 GB for 2-bit and up to 500GB for the full version. In a schema with a partial offloading to RAM requirements for VRAM could be less.
@Sumanth_077 Nice article! From my counts training even 7b model would require ~ 150k$ having A100 GPU with a rate 1$ per hour, is there a way to cut those expenses?