Update chatglm.h

intel · Mar 7, 2024 · a70ffc1 · a70ffc1
1 parent 77da4e2
commit a70ffc1
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/neural_speed/models/chatglm/chatglm.h b/neural_speed/models/chatglm/chatglm.h
@@ -26,7 +26,7 @@ enum chatglm_model {
 static const model_scratch chatglm_mem_req(int n_layers) {
   switch (n_layers) {
     case 28:
-      return {2048ull * MB, 2048ull * MB, 4096ull * MB};
+      return {4096ull * MB, 4096ull * MB, 8192ull * MB};
     // TODO(hengyu): add more variants besides 6B
     default:
       MODEL_ASSERT(false);