Remove redundant workers in Qwen3 Q8_0 FFN layers.

Author: orionpapadakis

src/main/java/org/beehive/gpullama3/tornadovm/layers/type/q8_0/Qwen3Q8_0FFNLayers.java

@@ -67,21 +67,12 @@ public Qwen3Q8_0FFNLayers(String taskGraphName, Qwen3State state, Qwen3TornadoWe
     public GridScheduler updateGridScheduler(GridScheduler gridScheduler) {
         WorkerGrid rmsNormWorker = WorkerGridFactory.createRmsNormWorker(config.dim(), state.localSize);
 
-        int matmulQGlobal = nEmbdHeadK * config.numberOfHeads() * LOCAL_WORK_GROUP_SIZE_ALLOC;
-        WorkerGrid matmulQRowMajorWorker = WorkerGridFactory.genericWorker(matmulQGlobal, LOCAL_WORK_GROUP_SIZE_ALLOC);
-
-        int matmulKVGlobal = nEmbdGqa * LOCAL_WORK_GROUP_SIZE_ALLOC;
-        WorkerGrid matmulKVRowMajorWorker = WorkerGridFactory.genericWorker(matmulKVGlobal, LOCAL_WORK_GROUP_SIZE_ALLOC);
-
         int qkRmsNormGroups = config.numberOfHeads() + config.numberOfKeyValueHeads();
         WorkerGrid qkRmsNormWorker = WorkerGridFactory.genericWorker(qkRmsNormGroups * nEmbdHead, nEmbdHead);
 
-        int h = config.numberOfHeads();
-        int ic = nEmbdHead / 2;
-        WorkerGrid ropeWorker = WorkerGridFactory.createRoPEWorker(h, nEmbdHead);
-        WorkerGrid copyToCachesWorker = WorkerGridFactory.genericWorker(nEmbdGqa, 128);
+        WorkerGrid ropeWorker = WorkerGridFactory.createRoPEWorker(config.numberOfHeads(), nEmbdHead);
         WorkerGrid parallelAttentionWorker = WorkerGridFactory.createAttentionWorker(config.numberOfHeads(), nEmbdHead);
-
+        // attn_output_proj worker (output projection)
         int matmul1Global = config.dim() * LOCAL_WORK_GROUP_SIZE_ALLOC;
         WorkerGrid matmul1Worker = WorkerGridFactory.genericWorker(matmul1Global, LOCAL_WORK_GROUP_SIZE_ALLOC);