@zephyr_z9 @Kimi_Moonshot I mean if Anthropic and others can why can't Kimi? It is a lot of parameters for the RAM even if in 4/8 bit, you'd expect it to be a multiple of v4 etc both for active parameters & for total parameters
@EMostaque @Kimi_Moonshot 60B-65B active They are generating at least 75%-85% margins on it
Guardrails removed spam, off-topic, unclear, or duplicate replies.
Ask a question below.
Published answers will appear here.
@zephyr_z9 @Kimi_Moonshot I mean if Anthropic and others can why can't Kimi? It is a lot of parameters for the RAM even if in 4/8 bit, you'd expect it to be a multiple of v4 etc both for active parameters & for total parameters
@EMostaque @Kimi_Moonshot 60B-65B active They are generating at least 75%-85% margins on it
Guardrails removed spam, off-topic, unclear, or duplicate replies.
Ask a question below.
Published answers will appear here.
