10-03-2026, 06:53 AM
Tried to reproduce it just now. On 5090 with vLLM, I saw basically the same behavior. The surprise was that memory pressure showed up before throughput collapsed.
|
NVFP4 / INT4 tradeoffs for Mistral
|
|
« Next Oldest | Next Newest »
|
| Messages In This Thread |
|
NVFP4 / INT4 tradeoffs for Mistral - by mangotinker73 - 09-29-2026, 01:18 AM
Re: NVFP4 / INT4 tradeoffs for Mistral - by mintstream9 - 09-30-2026, 01:38 AM
Re: NVFP4 / INT4 tradeoffs for Mistral - by rocmstack25 - 09-30-2026, 04:24 AM
Re: NVFP4 / INT4 tradeoffs for Mistral - by signaltrack44 - 09-30-2026, 04:46 AM
Re: NVFP4 / INT4 tradeoffs for Mistral - by thetashed24 - 09-30-2026, 05:16 AM
Re: NVFP4 / INT4 tradeoffs for Mistral - by vectorport - 10-01-2026, 05:41 PM
Re: NVFP4 / INT4 tradeoffs for Mistral - by cudapoint - 10-01-2026, 06:49 PM
Re: NVFP4 / INT4 tradeoffs for Mistral - by mintstream9 - 10-03-2026, 06:53 AM
Re: NVFP4 / INT4 tradeoffs for Mistral - by orbitgarden - 10-03-2026, 11:20 AM
Re: NVFP4 / INT4 tradeoffs for Mistral - by thetatinker77 - 10-03-2026, 12:00 PM
Re: NVFP4 / INT4 tradeoffs for Mistral - by cacheplane98 - 10-03-2026, 08:40 PM
Re: NVFP4 / INT4 tradeoffs for Mistral - by batchwire - 10-03-2026, 10:12 PM
|
Users browsing this thread: 1 Guest(s)