llama : add SKIP_KQ_KQV option

common/common.cpp

@@ -632,7 +632,6 @@ bool gpt_params_parse(int argc, char ** argv, gpt_params & params) {
         process_escapes(params.prompt);
         process_escapes(params.input_prefix);
         process_escapes(params.input_suffix);
-        process_escapes(sparams.cfg_negative_prompt);
         for (auto & antiprompt : params.antiprompt) {
             process_escapes(antiprompt);
         }

examples/main/main.cpp

@@ -761,9 +761,6 @@ int main(int argc, char ** argv) {
                         n_consumed = embd_inp.size();
                         embd_inp.insert(embd_inp.end(), inp_pfx.begin(), inp_pfx.end());
                     }
-                    if (params.escape) {
-                        process_escapes(buffer);
-                    }
 
                     const auto line_pfx = ::llama_tokenize(ctx, params.input_prefix, false, true);
                     const auto line_inp = ::llama_tokenize(ctx, buffer,              false, false);

ggml.c

@@ -16602,6 +16602,10 @@ static void ggml_compute_forward_cross_entropy_loss_back(
 static void ggml_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * tensor) {
     GGML_ASSERT(params);
 
+    if (tensor->op == GGML_OP_NONE) {
+        return;
+    }
+
 #ifdef GGML_USE_CUBLAS
     bool skip_cpu = ggml_cuda_compute_forward(params, tensor);
     if (skip_cpu) {

llama.cpp

@@ -5815,6 +5815,33 @@ static struct ggml_cgraph * llama_build_graph(
             GGML_ASSERT(false);
     }
 
+#if 1
+    for (int i = 0; i < result->n_nodes; ++i) {
+        struct ggml_tensor * node = result->nodes[i];
+        if (getenv("SKIP_KQ_ALL")) {
+            if (
+                    strcmp(node->name, "KQ")  == 0 ||
+                    strcmp(node->name, "KQ_scaled") == 0 ||
+                    strcmp(node->name, "KQ_masked") == 0 ||
+                    strcmp(node->name, "KQ_soft_max") == 0 ||
+                    strcmp(node->name, "KQV") == 0 ||
+                    false) {
+                //printf("skipping %s\n", dst->name);
+                node->op  = GGML_OP_NONE;
+            }
+        }
+        if (getenv("SKIP_KQ_KQV")) {
+            if (
+                    strcmp(node->name, "KQ")  == 0 ||
+                    strcmp(node->name, "KQV") == 0 ||
+                    false) {
+                //printf("skipping %s\n", dst->name);
+                node->op  = GGML_OP_NONE;
+            }
+        }
+    }
+#endif
+
     return result;
 }