We just hit 300 members today—woke up to 100+ new people joining overnight. Come join these top DIY hackers using extreme optimization to make old hardware run Qwen3.8-Flash and GLM5.3-Flash. 🚀xitter.cf/i/chat/group_join/g209…

ModelScope hasn’t synced the GSQ-RCO GGUF for DeepSeek-V4.1-Flash-1017 yet, so I’m downloading from three other mirrors at the same time—and the speed is just as good as ModelScope’s.

Strata+DeepSeek🔥

My new power dashboard? I love it. 📊

Quick tip for the new Strata build: don’t forget --conversation-cache-mib 0 to disable conversation caching. You’ll lose a little prefill speed—just a heads-up.

We’re hitting 400 members this week! Join us and let’s rock local LLMs together. 🚀

China actually has plenty of homegrown inference GPUs, but prices haven’t dropped as expected—and they’re still not great for local LLMs. Literally one big reason: no CUDA. Without the ecosystem—PyTorch, kernels, quants, llama.cpp/Strata support—local LLM tinkering gets way harder.

Oops—Strata + DFlash still doesn’t support multi-GPU yet. Guess I’ll wait for the next release. 😅

Downloading, love this dude🍻🍻🍻

DeepSeek V4.1 Flash “cheated” by checking Qwen’s previous work from memory, then ate up the 70 minutes I’d planned to sleep—plus $3. But I forgive you. The quality won me over. Take my knees—SOTA King! 🙇

Why 2T only cost $80~200 now? Not making any sense. Last year I paid $120 for one.

Like I said, new models come and new models come go—we all end up running only Qwen locally. DeepSeek V4.1 Flash is ridiculously slow on my V100 rig (256GB RAM + 96GB VRAM).

Qwen3.8-Flash-Next GSQ-RCO-IQ3_S vs UD Q4_K_XL—GSQ-RCO-IQ3_S(left) is 1.3x decode faster but both ran in 32 minutes, but UD Q4_K_XL(right) clearly wins on detail.

github.com/NVIDIA/flare-loca… how about V100?

Excuse me? Where did you even get Qwen4??

4× V100 96GB VRAM test on ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S: Strata is far ahead of llama.cpp+DFlash—9.05× prefill, 1.67× decode. Can’t wait for Strata + DFlash + this model. And the quant quality is f***ing good—on par with Q6 so far. Gonna run more tasks on it.

OMG, 4000 TPS prefill. I’ve never seen such speed on my V100 rig.😍

Qwen3.8-Flash IQ4 + DFlash2 + llama.cpp: decode up to 28% faster 🚀