fix: sampler vocab size. (#241)

Author: b4rtaz

src/app.cpp

@@ -234,9 +234,9 @@ void runInferenceApp(AppCliArgs *args, void (*handler)(AppInferenceContext *cont
 
     Tokenizer tokenizer(args->tokenizerPath);
     if (tokenizer.vocabSize != header.vocabSize)
-        printf("Tokenizer vocab size does not match the model vocab size: %d != %d\n", tokenizer.vocabSize, header.vocabSize);
+        printf("Tokenizer vocab size (%d) does not match the model vocab size (%d)\n", tokenizer.vocabSize, header.vocabSize);
 
-    Sampler sampler(header.vocabSize, args->temperature, args->topp, args->seed);
+    Sampler sampler(tokenizer.vocabSize, args->temperature, args->topp, args->seed);
 
     LlmNet net = buildLlmNet(&header, nNodes, args->nBatches);
     std::unique_ptr<LlmNet, void(*)(LlmNet *)> netPtr(&net, releaseLlmNet);

src/nn/nn-vulkan-test.cpp

@@ -421,7 +421,6 @@ void testCast_F32_Q80() {
             for (NnUint i = 0; i < N_BATCHES * dim; i++) {
                 const float expectedV = (float)(i + 1);
                 const float change = (yF32[i] - expectedV) / expectedV;
-                printf("cast %f %f\n", expectedV, yF32[i]);
                 assertFloat(i, change, 0.0, 0.009f);
             }
             printOk("testCast_F32_Q80");