perplexity only uses pp_threads

netrunnereve · netrunnereve · commit d854348992fe · 2023-08-08T21:30:12.000-04:00
diff --git a/examples/perplexity/perplexity.cpp b/examples/perplexity/perplexity.cpp
@@ -66,7 +66,7 @@ void perplexity(llama_context * ctx, const gpt_params & params) {
                 tokens[batch_start] = llama_token_bos();
             }
 
-            if (llama_eval(ctx, tokens.data() + batch_start, batch_size, j * n_batch, params.n_threads, params.n_threads)) {
+            if (llama_eval(ctx, tokens.data() + batch_start, batch_size, j * n_batch, params.n_threads, params.pp_threads)) {
                 fprintf(stderr, "%s : failed to eval\n", __func__);
                 return;
             }
@@ -337,8 +337,8 @@ int main(int argc, char ** argv) {
     // print system information
     {
         fprintf(stderr, "\n");
-        fprintf(stderr, "system_info: n_threads = %d / %d | %s\n",
-                params.n_threads, std::thread::hardware_concurrency(), llama_print_system_info());
+        fprintf(stderr, "system_info: pp_threads = %d / %d | %s\n",
+                params.pp_threads, std::thread::hardware_concurrency(), llama_print_system_info());
     }
 
     if (params.hellaswag) {

Original file line number	Diff line number	Diff line change
`@@ -66,7 +66,7 @@ void perplexity(llama_context * ctx, const gpt_params & params) {`
`66`	`66`	`tokens[batch_start] = llama_token_bos();`
`67`	`67`	`}`
`68`	`68`
`69`		`- if (llama_eval(ctx, tokens.data() + batch_start, batch_size, j * n_batch, params.n_threads, params.n_threads)) {`
	`69`	`+ if (llama_eval(ctx, tokens.data() + batch_start, batch_size, j * n_batch, params.n_threads, params.pp_threads)) {`
`70`	`70`	`fprintf(stderr, "%s : failed to eval\n", __func__);`
`71`	`71`	`return;`
`72`	`72`	`}`
`@@ -337,8 +337,8 @@ int main(int argc, char ** argv) {`
`337`	`337`	`// print system information`
`338`	`338`	`{`
`339`	`339`	`fprintf(stderr, "\n");`
`340`		`- fprintf(stderr, "system_info: n_threads = %d / %d \| %s\n",`
`341`		`- params.n_threads, std::thread::hardware_concurrency(), llama_print_system_info());`
	`340`	`+ fprintf(stderr, "system_info: pp_threads = %d / %d \| %s\n",`
	`341`	`+ params.pp_threads, std::thread::hardware_concurrency(), llama_print_system_info());`
`342`	`342`	`}`
`343`	`343`
`344`	`344`	`if (params.hellaswag) {`