@nahcrof Hi! Just wondering if it’s possible to know which exact quant the Q8_0 version of DeepSeek-V4-Flash-0731 is based on — whether it’s the Unsloth one (UD-Q8_K_XL) or a conversion from somewhere else (like Preyazz or custom).
No worries if it’s not clear, just curious. Thanks! 🙏
@nahcrof Will this be permanent? I understand it’s quantized to Q8. Hopefully they can improve the input cache rate (or maybe they already did). Either way, good prices to try it out.
DeepSeek V4 Flash on ClinePass feels noticeably less intelligent to me — as if it were quantized. On two occasions it completely froze in the middle of the agent’s execution, something that has never happened to me with the official DeepSeek. I also saw it reasoning in a garbled, scribbly way.
@elonmusk Un flag o algo para configurar en el modelo para indicarle que no tiene visión por ejemplo cuando uso DeepSeek v4 flash me sale muchos errores porque intenta procesar imágenes y la sesión de bloquea salen puros 400
@veggie_eric Could you offer a better usage rate for https://t.co/hiOBg8uczo to X premium users? I can only make 20 queries every 2 hours. On X's Grok, I can use it much more.