miami office
bneo.xyz · 214 words · saved by 1 readers
writing daily
October 6, 2025 · 1 min read i got room service. acai bowl and some sausages and a smoothie. had a chat with e about the plan and updates. the goal was to speed up and try to fit 70b in one h100 instead of two. tried out online dynamic quantization with vllm (fp8). it was actually slower than before and also still required the same amount of memory. pointless. next explored qlora training with bitsandbytes. kept facing 0.0 grad_norm issues and tried to debug the entire day and couldn't figure out why. a problem for another day to solve. at night all of us went to the rooftop and everyone…
saved by
related reading
- end of 2025bneo.xyz
- choosing - by vincent huang - a slice of my mindmindslice.substack.com
- Vincent Huangvvhuang.com
- Reflections on 2025samuelalbanie.substack.com
- How I became a machine learning practitionerblog.gregbrockman.com
- blanket judgements and bridge opinions - by Sasha Hydrieshydrie.substack.com
- this might be the greatest city in the worldshydrie.substack.com
- SARAH CHIENG - Blogsarahchieng.weebly.com
- about - naklecha.comnaklecha.com
- Thoughts — Jason Weijasonwei.net
- Reflections on 2025benanderson.work
- My 2023 in NYC, and what I’m thinking about for 2024giansegato.com