shutdown #gpuapi Lambda server after succesful experiment playing around with GPU but it costs $1000/mo to keep a H100 running

split #gpuapi in API and worker with Redis queue to fix parallel threading issues that blocks new requests when images are generating #photoai

fix parallel Python threading issue #gpuapi

run Claude Code with --dangerously-skip-permissions first time to fix #gpuapi threading

fix #gpuapi can't send webhooks WHILE it's generating images

write blog about my crazy experience putting Claude Code on a Lambda GPU VPS and build #gpuapi x.com/levelsio/status/1951243…

manage to make FP8 quantized Flux model work with LoRa which means we have enough ram now to run parallel pipeline (more jobs at same time) #gpuapi

once again move #gpuapi files to the permanent storage

move entire GPU server to persistent storage so we don't lose it if server goes down #gpuapi

add #gpuapi dashboard to see live jobs

move #gpuapi files to permanent storage filesystem on Lambda

discover a GH200 could easily handle 16-32 image jobs parallel at same time and make it parallel #gpuapi

make #gpuapi async webhooks work and generate first images from #photoai interface with my own GPU

make #gpuapi work on #photoai from the interface and generate image

make #gpuapi GitHub repo and ask Claude Code to put everything in there by itself

reach Claude usage limits and upgrade #gpuapi

set up Nginx server on gpuapi.net #gpuapi to expose the H100 GPU and accept API requests and return photos generated

create gpuapi.net to try create my own #gpuapi with Claude Code

accidentally delete entire Docker on GPU server so we need to reinstall everything LOL #gpuapi

try generate Flux image with my LoRa on my GPU #gpuapi