Merge pull request #2 from jamorphy/wip-orpheus-tts

cleanup

Author: jsrgb

examples/tts/orpheus-tts.cpp

@@ -298,7 +298,7 @@ int main(int argc, char **argv) {
     params.model = params.vocoder.model;
     params.n_batch = 2;
 
-    params.embedding = true
+    params.embedding = true;
     // disable warmup, SNAC doesn't care about BOS or EOS tokens;
     params.warmup = false; 
 

src/llama-context.cpp

@@ -312,9 +312,7 @@ llama_context::llama_context(
 
         // reserve pp graph first so that buffers are only allocated once
         {
-            LLAMA_LOG_DEBUG("here 3\n");
             llama_ubatch ubatch_pp = { true, n_tokens, n_tokens / n_seqs, n_seqs, &token, nullptr, nullptr, nullptr, nullptr, nullptr};
-
             auto * gf = graph_init();
             graph_build(ctx_compute.get(), gf, ubatch_pp, LLM_GRAPH_TYPE_DEFAULT);
             if (!ggml_backend_sched_reserve(sched.get(), gf)) {

src/llama-model.cpp

@@ -1495,8 +1495,7 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
                 if (flags & TENSOR_NOT_REQUIRED) {
                     return nullptr;
                 }
-                return nullptr;
-                //throw std::runtime_error(format("missing tensor '%s'", tn.str().c_str()));
+                throw std::runtime_error(format("missing tensor '%s'", tn.str().c_str()));
             }
 
             // some models use the token embedding tensor as the output, but since these are used in different layers and with different ops
@@ -1591,8 +1590,6 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
                     return t;
                 }
             }
-            fprintf(stderr, "create_tensor: Creating '%s' with ne=[%ld, %ld, %ld]\n",
-        tn_str.c_str(), ne.begin()[0], ne.begin()[1], ne.begin()[2]);
             return ml.create_tensor(ctx, tn, ne, flags);
         };