Florian Laurent
@MasterScrat
Followers
2K
Following
18K
Statuses
4K
Building ⚡️https://t.co/YQusEdDrpc - text2image model training on TPUs
Switzerland
Joined May 2009
@ludwigABAP "showing all CoT leads to competitive distillation"
@Teknium1 They said 3 days ago they had no plan to show them for real "We're working on showing a bunch more than we show today—that will be very very soon. TBD on all, showing all CoT leads to competitive distillation"
0
0
1
RT @ComfyUI: "Mapping made with ComfyUI Zero prompt. Only control 👌 Thanks to SUPERBIEN for giving me the opportunity to work on this Not…
0
60
0
@Teknium1 They said 3 days ago they had no plan to show them for real "We're working on showing a bunch more than we show today—that will be very very soon. TBD on all, showing all CoT leads to competitive distillation"
1
0
30
RT @Dorialexander: In case it interests anyone, I managed to set up a workable version of @willccbb script on Colab. Replace llama 1B with…
0
73
0
03-mini-2025-01-31 its final result is different from what GPT-4o came up with Asking for details, it's referring to "different lists", "commonly accepted answer" as if summarizing the work of a team
New favorite eval question: How many countries have more vowels than consonants in the name of their capital city? GPT-4: "There are 19 countries with capital cities that have more vowels than consonants in their names." It listed all capitals, then wrote a Python script to do the calculation. GPT-4o: "Several countries have capital cities whose names contain more vowels than consonants. Here are some examples: ..." 🙄 QwQ: thinks forever
0
0
0
RT @andersonbcdefg: looks like RL is the future. stop grinding leetcode, kiddo. start grinding cartpole-v1
0
23
0
@multimodalart @linoy_tsaban amazing! it's been gone too long. it was the first patreon i ever joined
0
0
1
RT @hkproj: Many people leaving everything and grind RL because it’s the trend. Others going all in on CUDA. Same people who will start gri…
0
61
0
@multimodalart prompt following and light is good, but overall not as good as Flux Dev imo (no cherrypicking for either):
0
0
2
somehow still up after 13y
0
0
0