debug

Author: kirklandsign

extension/llm/runner/llm_runner_helper.cpp

@@ -111,9 +111,9 @@ get_llm_metadata(tokenizers::Tokenizer* tokenizer, Module* module) {
   if (!method_names.count(llm::kMaxSeqLen)) {
     ET_LOG(
         Error,
-        "Required metadata method %s not found in model",
+        "Required metadata method %s not found in model. Bypass",
         llm::kMaxSeqLen);
-    return ::executorch::runtime::Error::InvalidArgument;
+    // return ::executorch::runtime::Error::InvalidArgument;
   }
 
   for (auto& pair : metadata) {

extension/llm/runner/multimodal_prefiller.cpp

@@ -37,6 +37,8 @@ MultimodalPrefiller::MultimodalPrefiller(
 Result<uint64_t> MultimodalPrefiller::prefill(
     const MultimodalInput& input,
     int64_t& start_pos) {
+      ET_LOG(Error, "Here 000000000000000000000000000000000000000000000000000000000000000000 %d", (int) (start_pos));
+      ET_LOG(Error, "Here 000000000000000000000000000000000000000000000000000000000000000000 %s", input.get_text().c_str());
   // 1. Run encoder model.
   ::executorch::runtime::EValue encoder_output;
   if (input.is_image()) {
@@ -73,12 +75,14 @@ Result<uint64_t> MultimodalPrefiller::prefill(
     auto& text = input.get_text();
     std::vector<uint64_t> tokens =
         ET_UNWRAP_TOKENIZER(tokenizer_->encode(text));
+    ET_LOG(Error, "Here 666666666666666666666666666666666666666666666666666666666666666666666666666666");
 
     auto text_tensor = executorch::extension::from_blob(
         tokens.data(),
         {1, static_cast<aten::SizesType>(tokens.size())},
         ::executorch::aten::ScalarType::Long);
 
+    ET_LOG(Error, "Here 555555555555555555555555555555555555555555555555555555555555555555555555555");
     // Run text encoder (token embeddings)
     auto token_embedding_outputs =
         ET_UNWRAP(module_->execute(kTokenEmbeddingMethod, text_tensor));
@@ -89,6 +93,7 @@ Result<uint64_t> MultimodalPrefiller::prefill(
     // For any other input types, return error
     return ::executorch::runtime::Error::NotSupported;
   }
+    ET_LOG(Error, "Here 000000000000000000000000000000000000000000000000000000000000000000000000000");
 
   // 2. Run decoder model for prefill.
   // `cache_position` goes from start_pos to start_pos + encoder_output.size(1).
@@ -107,6 +112,7 @@ Result<uint64_t> MultimodalPrefiller::prefill(
       cache_positions.data(),
       {static_cast<int>(seq_len)},
       executorch::aten::ScalarType::Long);
+    ET_LOG(Error, "Here 111111111111111111111111111111111111111111111111111111111111111111111111111111111111111");
   auto prefill_result = module_->execute(
       kTextModelMethod, {cache_position_tensor, encoder_output});
   if (prefill_result.error() != ::executorch::runtime::Error::Ok) {
@@ -121,6 +127,7 @@ Result<uint64_t> MultimodalPrefiller::prefill(
     return ::executorch::runtime::Error::InvalidState;
   }
   auto outputs_res = prefill_outputs[0].toTensor();
+    ET_LOG(Error, "Here 222222222222222222222222222222222222222222222222222222222222222222222222222222");
 
   // Update start_pos, tracking the current cache position.
   start_pos += seq_len;

extension/llm/runner/multimodal_runner.cpp

@@ -57,9 +57,9 @@ Error MultimodalRunner::load() {
 // Don't print with the same priority during warmup
 #define RUNNER_ET_LOG(warmup, format, ...) \
   if (warmup) {                            \
-    ET_LOG(Debug, format, __VA_ARGS__);    \
+    ET_LOG(Error, format, __VA_ARGS__);    \
   } else {                                 \
-    ET_LOG(Info, format, __VA_ARGS__);     \
+    ET_LOG(Error, format, __VA_ARGS__);     \
   }
 
 Error MultimodalRunner::generate(
@@ -104,16 +104,20 @@ Error MultimodalRunner::generate(
 
   uint64_t prefill_next_token = 0;
   // Process multimodal inputs in order
+  ET_LOG(Error, "0000000000000000000000000000000000000000000000000000SIZE%d", inputs.size());
   for (const MultimodalInput& input : inputs) {
+  ET_LOG(Error, "00000000000000000000000000000000123321451345143100");
     prefill_next_token = ET_UNWRAP(multimodal_prefiller_->prefill(input, pos_));
   }
+  ET_LOG(Error, "1111111111111111111111111111111111111111111111111111");
 
   stats_->first_token_ms = time_in_ms();
   stats_->prompt_eval_end_ms = time_in_ms();
   stats_->num_prompt_tokens = pos_;
 
   wrapped_callback(ET_UNWRAP_TOKENIZER(
       tokenizer_->decode(prefill_next_token, prefill_next_token)));
+  ET_LOG(Info, "2222222222222222222222222222222222222222222222222222");
 
   RUNNER_ET_LOG(
       config.warming,