Commit 9d5187eb authored by zhuwenwen's avatar zhuwenwen
Browse files

解决PA 大batch,32k,崩溃问题

parent 95d58c0d
...@@ -965,7 +965,7 @@ void get_numberthread_and_reuse_kv_v2(int& num_thread,int& reusekv,int& PARTITIO ...@@ -965,7 +965,7 @@ void get_numberthread_and_reuse_kv_v2(int& num_thread,int& reusekv,int& PARTITIO
) )
max_num_partitions=1; max_num_partitions=1;
int blocks=max_num_partitions*batchsize*qheads; int blocks=max_num_partitions*batchsize*qheads;
if(device_name=="gfx928"){ if(device_name=="gfx928"||batchsize>100){
if(batchsize*qheads>1024&&max_seq_len>=2000){ if(batchsize*qheads>1024&&max_seq_len>=2000){
max_num_partitions=1; max_num_partitions=1;
if(max_seq_len<2000)reusekv=8; if(max_seq_len<2000)reusekv=8;
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment