是否有支持deepseek-r1 671B fp16的打算？deepseek-r1 671B q8_0需要多大的DRAM,VRAM？

[ProgramerX]

我想请教几个问题：
（1）ktransformers是否支持deepseek-r1 671B fp16，如果支持，需要多大的DRAM,VRAM?如果不支持，后续是否有支持的打算？
（2）ktransformers运行deepseek-r1 671B q8_0需要多大的DRAM,VRAM?

[les712]

671B fp16 显存+内存至少1.4T，不要想了。能跑Q4量化已经很不错了

[CYSTEV-chn]

deepseek-r1 671B q8_0, DRAM 768G，VRAM 24G,run ok!

[AYOUYI]

Nice setup! How many tokens/sec are you getting with that 24GB VRAM and 8-bit quantization?

[voipmonitor]

deepseek-r1 671B q8_0, DRAM 768G，VRAM 24G,run ok!

Can you please tell me the exact ktranformars cmd / parameters how did you run it on your setup?

[gitCky]

Can you please tell me the command line you used to launch DeepSeek-R1 671B q8_0? My model weights stop loading at layer 47, and the log window freezes. Can you help me with this?