bgeneto

[bgeneto @netobge]

@superalesha You missed Intel AutoRound 4-bit, it provides better performance for your RTX 3090... I think it's worth testing!

[bgeneto @netobge]

@elivieira Impeachment nele já!

[bgeneto @netobge]

@opencode Is DeepSeek Flash v4 0731 actually ready for production? Would anyone trust it for medium-to-heavy coding tasks?

[bgeneto @netobge]

@alecqfong Very cheap way to get those speeds indeed... 🤭

[bgeneto @netobge]

@flerimeja @LuminaXspace @cursor_ai What's the purpose of that 3D view of your codebase?

[bgeneto @netobge]

@Alibaba_Qwen So we can expect something like Qwen3.8 27B and Qwen3.8 35B A3B for consumer GPUs?

[bgeneto @netobge]

@leerob Thx... It was draining really fast, faster then normal. Grok 4.5 should be computed as First-party model usage now after the acquisition, right?

[bgeneto @netobge]

@superbiche @sudoingX 3090 can do near 260 tps gen with 35B A3B with vLLM, Intel AutoRound 4-bit, fp8 kv-cache without any practical (<1%) accuracy loss

[bgeneto @netobge]

@cursor_ai @cursor_ai How many parameters are in the new Grok 4.5? Does it run as fast as Composer 2.5?

[bgeneto @netobge]

@Psicolicious @UnslothAI FP8 kv cache in vLLM, disable video and tweak image settings (lower resolution supported). Can do 262k context with text-only easily. Lost only 1% (mean) in all benchmarks with this amazing setup.

[bgeneto @netobge]

@Tech2Wild Here it is... ready to run docker compose file:
https://t.co/eVG57NuPCk

[bgeneto @netobge]

@acadictive This 30x price increase should be considered illegal by any means!

[bgeneto @netobge]

@acadictive Burned in the first day. You are a hero! @GitHubCopilot new way of counting tokens is completely unfair.

[bgeneto @netobge]

@TheAhmadOsman Qwen3.7 Max is not open-source/open-weights, right? So OP is probably right, Let's see how M3 stacks against Kimi though

[bgeneto @netobge]

@danieltvela @The_Only_Signal It really is... but then what would we call Qwen3.6 35B A3B (called Qwen3.6 Flash by Alibaba) running at 160tps on a single rtx 3090 with 262k context in vLLM vs 37 tps 27B then?

[bgeneto @netobge]

@leo_linsky @The_Only_Signal Interesting to see the Qwen3.6 flash, a 35B A3B MoE, ahead of 27B dense and all the hype going to the much slower 27B

[bgeneto @netobge]

@CarlosZarattini Deixa ver se eu entendi sua posição: então o dado que desmonta essa mentira de vez diz que menos de 3% dos beneficiários trabalham de carteira assinada... Puxa, 3% é realmente um desmonte...

[bgeneto @netobge]

@cjzafir Have you published your model? Where?