DeepSeek V3 (GitHub)
· China Labs
support scale_fmt=ue8m0 (#964)
support scale_fmt=ue8m0 (#964) * support scale_fmt=ue8m0 * keep improving Signed-off-by: youkaichao * keep improving Signed-off-by: youkaichao * add clamp min of 1e-4 Signed-off-by: youkaichao * rename config Signed-off-by: youkaichao