MTP layers missing?
#1
by g-a-b-y - opened
Hello,
I noticed that the MTP/NextN weights appear to be missing from the checkpoint.
Also is the FP8 kv cache calibrated?
Ping @shengliangx
Hello @g-a-b-y this model's MTP is the 45th layer. It is in the checkpoint. The MTP layer is not quantized in this checkpoint.
g-a-b-y changed discussion status to closed