LocalAiLab

[LocalAiLab @LocalAiLabq0ie]

@MiaAI_lab Try Qwen3.8 flash next using ssd loading instead of ram! With sglang you should be able to get high tok/s just step won't be that easy

[LocalAiLab @LocalAiLabq0ie]

@LeoBuilds_ @NovaXCode Muse Spark 1.3 is amazing

[LocalAiLab @LocalAiLabq0ie]

@MiaAI_lab @XiaomiMiMo They entered the game late, but better than never, let's give them 6 months time, if they can't achieve top frontier, then they out

[LocalAiLab @LocalAiLabq0ie]

@OrcaRouter How is the model only 7B?

[LocalAiLab @LocalAiLabq0ie]

@ItsmeAjayKV What concern me is that Chinese lab are silent for the past one or two months, ofc qwen came out and so on, but even Qwen3.8 max is not near lvl of opus 5.5 or fable5.1 or Astra! Usually their answer was fast or I'm impatient. Hope they beat closed models at any lvl

[LocalAiLab @LocalAiLabq0ie]

@MiaAI_lab 😂👍

[LocalAiLab @LocalAiLabq0ie]

@mehulmpt Add him to Strix, it does hack anywhere you want!

[LocalAiLab @LocalAiLabq0ie]

@tpacker @uday_devops Also it's not about one time production, the same software my evolves for years and needs to works perfectly fine. This spaghetti will breaks somewhere you never expect it

[LocalAiLab @LocalAiLabq0ie]

@tpacker @uday_devops Until it breaks, or new features break it or being hacked!

[LocalAiLab @LocalAiLabq0ie]

@uday_devops This is the best describing photo I've ever seen for this! Well done Sir 🫡

[LocalAiLab @LocalAiLabq0ie]

@solopribuilds Yes, it's bench maxed! I have many llm in access, opus 5.5 did hallucinate, while other llm (muse spark 1.3, glm5.3 flash, Qwen3.8 27B, kimi k3, gpt Astra) all solved the problem, it took them while to do it, but they did it, while claude did change the wrong stuff. same prompt

[LocalAiLab @LocalAiLabq0ie]

@letdarky Who check about Gemini? 😂

[LocalAiLab @LocalAiLabq0ie]

@vibeonX69 All 4 😂

[LocalAiLab @LocalAiLabq0ie]

@arena @AnthropicAI Bench maxed! I did test it and I have access to lot of models, it did hallucinate

[LocalAiLab @LocalAiLabq0ie]

@ahmedkhaild9510 @QwenDevs Unsloth fp8 version with comfy ui

[LocalAiLab @LocalAiLabq0ie]

Qwen image2.1 running locally on rtx4050 laptop with only 6gb of vram. @QwenDevs thank you guys

[LocalAiLab @LocalAiLabq0ie]

@S__2__2__2__S Unsloth fp8 version with comfy ui

[LocalAiLab @LocalAiLabq0ie]

Qwen image2.1 running locally on rtx4050 laptop with only 6gb of vram

[LocalAiLab @LocalAiLabq0ie]

@karanC_12 He scares of open source Ai runs locally

[LocalAiLab @LocalAiLabq0ie]

@Polymarket Wait for qwen4 27B to beat it with just download, unzip, run 😂

[LocalAiLab @LocalAiLabq0ie]

@Biggest There is no 35B MoE more! They announced what they want to release, you too late! Flash is MoE, but probably 100B param + 50B ngram