examples : fix build + compile warnings (close #1256)

examples/common.cpp

@@ -792,7 +792,7 @@ bool sam_params_parse(int argc, char ** argv, sam_params & params) {
     return true;
 }
 
-void sam_print_usage(int argc, char ** argv, const sam_params & params) {
+void sam_print_usage(int /*argc*/, char ** argv, const sam_params & params) {
     fprintf(stderr, "usage: %s [options]\n", argv[0]);
     fprintf(stderr, "\n");
     fprintf(stderr, "options:\n");

examples/lsp/lsp.cpp

@@ -324,12 +324,12 @@ json register_commandset(struct whisper_context * ctx, json jparams, std::vector
     commandset_list.push_back(cs);
     return json{{"index",index}};
 }
-json seek(struct whisper_context * ctx, audio_async &audio, json params) {
+json seek(struct whisper_context * /*ctx*/, audio_async & /*audio*/, json /*params*/) {
     // whisper_state has the pertinent offsets, but there also seem to be a large
     // number of scratch buffers that would prevent rewinding context in a manner similar to llama
     // I'll give this a another pass once everything else is implemented,
     // but for now, it's unsupported
-    throw json{
+    throw json {
         {"code", -32601},
             {"message", "Seeking is not yet supported."}
     };
@@ -412,7 +412,7 @@ void process_loop(struct whisper_context * ctx, audio_async &audio, const whispe
             jobqueue.pop_front();
             // send response
             std::string data = resp.dump(-1, ' ', false, json::error_handler_t::replace);
-            fprintf(stdout, "Content-Length: %d\r\n\r\n%s\n", data.length()+1, data.c_str());
+            fprintf(stdout, "Content-Length: %d\r\n\r\n%s\n", (int)data.length()+1, data.c_str());
             std::cout.flush();
 
         }

examples/main/main.cpp

@@ -260,7 +260,7 @@ std::string estimate_diarization_speaker(std::vector<std::vector<float>> pcmf32s
 
     return speaker;
 }
-void whisper_print_progress_callback(struct whisper_context * ctx, struct whisper_state * /*state*/, int progress, void * user_data) {
+void whisper_print_progress_callback(struct whisper_context * /*ctx*/, struct whisper_state * /*state*/, int progress, void * user_data) {
     int progress_step = ((whisper_print_user_data *) user_data)->params->progress_step;
     int * progress_prev  = &(((whisper_print_user_data *) user_data)->progress_prev);
     if (progress >= *progress_prev + progress_step) {
@@ -492,7 +492,7 @@ bool output_csv(struct whisper_context * ctx, const char * fname, const whisper_
     return true;
 }
 
-bool output_score(struct whisper_context * ctx, const char * fname, const whisper_params & params, std::vector<std::vector<float>> pcmf32s) {
+bool output_score(struct whisper_context * ctx, const char * fname, const whisper_params & /*params*/, std::vector<std::vector<float>> /*pcmf32s*/) {
     std::ofstream fout(fname);
     fprintf(stderr, "%s: saving output to '%s'\n", __func__, fname);
 

examples/talk-llama/llama.cpp

@@ -1164,7 +1164,7 @@ static bool llama_eval_internal(
     const llama_token * tokens,
             const int   n_tokens,
             const int   n_past,
-            const int   n_threads) {
+                  int   n_threads) {
 
     // enforce that the first token is BOS
     if (n_past == 0 && tokens[0] != llama_token_bos()) {
@@ -1190,6 +1190,8 @@ static bool llama_eval_internal(
     const int n_vocab = hparams.n_vocab;
     const int n_rot   = hparams.n_embd/hparams.n_head;
 
+    const float eps = 5e-6f; // TODO: take from hparams
+
     auto & mem_per_token = lctx.mem_per_token;
     auto & buf_compute   = lctx.buf_compute;
 
@@ -1204,7 +1206,7 @@ static bool llama_eval_internal(
     // for big prompts, if BLAS is enabled, it is better to use only one thread
     // otherwise, the threads are spin-lock waiting for the BLAS calls and are degrading the performance
     ggml_cgraph gf = {};
-    gf.n_threads = N >= 32 && ggml_cpu_has_blas() && !ggml_cpu_has_gpublas() ? 1 : n_threads;
+    n_threads = N >= 32 && ggml_cpu_has_blas() && !ggml_cpu_has_gpublas() ? 1 : n_threads;
 
     struct ggml_tensor * embd = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, N);
     ggml_set_name(embd, "embd");
@@ -1221,7 +1223,7 @@ static bool llama_eval_internal(
 
         // norm
         {
-            cur = ggml_rms_norm(ctx0, inpL);
+            cur = ggml_rms_norm(ctx0, inpL, eps);
 
             // cur = cur*attention_norm(broadcasted)
             cur = ggml_mul(ctx0, cur, model.layers[il].attention_norm);
@@ -1329,7 +1331,7 @@ static bool llama_eval_internal(
         {
             // norm
             {
-                cur = ggml_rms_norm(ctx0, inpFF);
+                cur = ggml_rms_norm(ctx0, inpFF, eps);
 
                 // cur = cur*ffn_norm(broadcasted)
                 cur = ggml_mul(ctx0, cur, model.layers[il].ffn_norm);
@@ -1367,7 +1369,7 @@ static bool llama_eval_internal(
     // norm
     {
 
-        inpL = ggml_rms_norm(ctx0, inpL);
+        inpL = ggml_rms_norm(ctx0, inpL, eps);
 
         // inpL = inpL*norm(broadcasted)
         inpL = ggml_mul(ctx0, inpL, model.norm);
@@ -1384,8 +1386,8 @@ static bool llama_eval_internal(
     //inpL = ggml_soft_max_inplace(ctx0, inpL);
 
     // run the computation
-    ggml_build_forward_expand(&gf, inpL);
-    ggml_graph_compute       (ctx0, &gf);
+    ggml_build_forward_expand  (&gf, inpL);
+    ggml_graph_compute_with_ctx(ctx0, &gf, n_threads);
 
 #ifdef GGML_PERF
     // print timing information per ggml operation (for debugging purposes)
@@ -2488,8 +2490,7 @@ int llama_apply_lora_from_file_internal(struct llama_context * ctx, const char *
             }
 
             struct ggml_cgraph gf = ggml_build_forward(r);
-            gf.n_threads = n_threads;
-            ggml_graph_compute(lora_ctx, &gf);
+            ggml_graph_compute_with_ctx(lora_ctx, &gf, n_threads);
 
             // we won't need these tensors again, reset the context to save memory
             ggml_free(lora_ctx);
@@ -2635,7 +2636,6 @@ size_t llama_copy_state_data(struct llama_context * ctx, uint8_t * dst) {
 
             ggml_context * cpy_ctx = ggml_init({ sizeof(buffer), buffer, /* no_alloc */ true });
             ggml_cgraph gf{};
-            gf.n_threads = 1;
 
             ggml_tensor * kout3d = ggml_new_tensor_3d(cpy_ctx, kv_self.k->type, n_embd, kv_ntok, n_layer);
             kout3d->data = out;
@@ -2655,7 +2655,7 @@ size_t llama_copy_state_data(struct llama_context * ctx, uint8_t * dst) {
 
             ggml_build_forward_expand(&gf, ggml_cpy(cpy_ctx, k3d, kout3d));
             ggml_build_forward_expand(&gf, ggml_cpy(cpy_ctx, v3d, vout3d));
-            ggml_graph_compute(cpy_ctx, &gf);
+            ggml_graph_compute_with_ctx(cpy_ctx, &gf, 1);
 
             ggml_free(cpy_ctx);
         }
@@ -2743,7 +2743,6 @@ size_t llama_set_state_data(struct llama_context * ctx, uint8_t * src) {
 
             ggml_context * cpy_ctx = ggml_init({ sizeof(buffer), buffer, /* no_alloc */ true });
             ggml_cgraph gf{};
-            gf.n_threads = 1;
 
             ggml_tensor * kin3d = ggml_new_tensor_3d(cpy_ctx, kv_self.k->type, n_embd, kv_ntok, n_layer);
             kin3d->data = (void *) inp;
@@ -2763,7 +2762,7 @@ size_t llama_set_state_data(struct llama_context * ctx, uint8_t * src) {
 
             ggml_build_forward_expand(&gf, ggml_cpy(cpy_ctx, kin3d, k3d));
             ggml_build_forward_expand(&gf, ggml_cpy(cpy_ctx, vin3d, v3d));
-            ggml_graph_compute(cpy_ctx, &gf);
+            ggml_graph_compute_with_ctx(cpy_ctx, &gf, 1);
 
             ggml_free(cpy_ctx);
         }

examples/talk-llama/talk-llama.cpp

@@ -649,7 +649,10 @@ int main(int argc, char ** argv) {
                 }
 
                 text_to_speak = ::replace(text_to_speak, "\"", "");
-                system((params.speak + " " + std::to_string(voice_id) + " \"" + text_to_speak + "\"").c_str());
+                int ret = system((params.speak + " " + std::to_string(voice_id) + " \"" + text_to_speak + "\"").c_str());
+                if (ret != 0) {
+                    fprintf(stderr, "%s: failed to speak\n", __func__);
+                }
 
                 audio.clear();
 

examples/talk/gpt-2.cpp

@@ -191,9 +191,9 @@ bool gpt2_model_load(const std::string & fname, gpt2_model & model, gpt_vocab &
     // create the ggml context
     {
         struct ggml_init_params params = {
-            .mem_size   = ctx_size,
-            .mem_buffer = NULL,
-            .no_alloc   = false,
+            /*.mem_size   =*/ ctx_size,
+            /*.mem_buffer =*/ NULL,
+            /*.no_alloc   =*/ false,
         };
 
         model.ctx = ggml_init(params);

examples/talk/talk.cpp

@@ -349,7 +349,10 @@ int main(int argc, char ** argv) {
                 gpt2_set_prompt(ctx_gpt, prompt_base.c_str());
 
                 text_to_speak = ::replace(text_to_speak, params.person + ": ", "");
-                system((params.speak + " " + std::to_string(voice_id) + " \"" + text_to_speak + "\"").c_str());
+                int ret = system((params.speak + " " + std::to_string(voice_id) + " \"" + text_to_speak + "\"").c_str());
+                if (ret != 0) {
+                    fprintf(stderr, "%s: system() failed!\n", __func__);
+                }
 
                 audio.clear();