Merge pull request #527 from ufownl:feature/gemma2_secondary_eos

PiperOrigin-RevId: 740327973

gemma/configs.cc

@@ -35,6 +35,8 @@ static ModelConfig ConfigBaseGemmaV2() {
   ModelConfig config = ConfigNoSSM();
   config.att_cap = 50.0f;
   config.final_cap = 30.0f;
+  config.eos_id = 1;
+  config.secondary_eos_id = 107;
   return config;
 }
 

gemma/gemma-inl.h

@@ -1427,7 +1427,7 @@ void GenerateT(const ModelWeightsStorage& model, Activations& activations,
   // Sanity check: prompts should not be empty, nor start with EOS.
   for (size_t query_idx = 0; query_idx < queries_prompt.size(); ++query_idx) {
     const PromptTokens& prompt = queries_prompt[query_idx];
-    HWY_ASSERT(prompt.size() != 0 && !model.Config().IsEOS(prompt[0]));
+    HWY_ASSERT(prompt.size() != 0 && prompt[0] != runtime_config.eos_id);
   }
 
   const size_t num_queries = queries_prompt.size();
@@ -1615,4 +1615,4 @@ void GenerateImageTokens(  // NOLINT(misc-definitions-in-headers)
 }  // namespace gcpp
 HWY_AFTER_NAMESPACE();
 
-#endif  // THIRD_PARTY_GEMMA_CPP_GEMMA_GEMMA_INL_H_
+#endif  // THIRD_PARTY_GEMMA_CPP_GEMMA_GEMMA_INL_H_

gemma/run.cc

@@ -85,7 +85,6 @@ void ReplGemma(Gemma& model, KVCache& kv_cache, const AppArgs& app,
   size_t abs_pos = 0;                     // across turns
   size_t tokens_generated_this_turn = 0;  // differentiates prefill from reply
   size_t prompt_size = 0;
-  bool end_of_turn_seen = false;
 
   std::mt19937 gen;
   InitGenerator(args, gen);
@@ -118,12 +117,6 @@ void ReplGemma(Gemma& model, KVCache& kv_cache, const AppArgs& app,
   // callback function invoked for each generated token.
   auto stream_token = [&](int token, float) {
     ++abs_pos;
-    if (model.GetModelConfig().IsEOS(token)) {
-      if (app.verbosity >= 2) {
-        std::cout << "\n[ End ]\n";
-      }
-      return true;
-    }
     const bool in_prompt = tokens_generated_this_turn < prompt_size;
     const bool first_response_token = tokens_generated_this_turn == prompt_size;
     ++tokens_generated_this_turn;
@@ -132,6 +125,11 @@ void ReplGemma(Gemma& model, KVCache& kv_cache, const AppArgs& app,
         std::cerr << "." << std::flush;
       }
       return true;
+    } else if (model.GetModelConfig().IsEOS(token)) {
+      if (app.verbosity >= 2) {
+        std::cout << "\n[ End ]\n";
+      }
+      return true;
     }
     std::string token_text;
     HWY_ASSERT(model.Tokenizer().Decode(std::vector<int>{token}, &token_text));
@@ -141,13 +139,6 @@ void ReplGemma(Gemma& model, KVCache& kv_cache, const AppArgs& app,
         std::cout << "\n\n";
       }
     }
-    if (token_text == "<end_of_turn>") {
-      // We don't want to show the <end_of_turn> token to the user.
-      // We also need to remember that we've seen it, so that we can rewind
-      // abs_pos appropriately. We expect EOS as the next token.
-      end_of_turn_seen = true;
-      return true;
-    }
     std::cout << token_text << std::flush;
     return true;
   };
@@ -233,13 +224,6 @@ void ReplGemma(Gemma& model, KVCache& kv_cache, const AppArgs& app,
       HWY_ASSERT(abs_pos > 0);
       abs_pos--;
     }
-    if (end_of_turn_seen && abs_pos > 0) {
-      // If we have seen an end_of_turn token, we need to rewind abs_pos by one
-      // more, because we will prepend it again to the prompt in
-      // WrapAndTokenize.
-      abs_pos--;
-    }
-    end_of_turn_seen = false;
   }
 }