apply copilot

Signed-off-by: fishbell <[email protected]>

Author: songbell

src/cpp/src/continuous_batching/pipeline.cpp

@@ -60,7 +60,7 @@ extract_eagle3_mode_from_config(ov::AnyMap& config, const std::filesystem::path&
             // If you wish to use different layers, provide the "hidden_layers_list" parameter in the config.
             eagle_rt_info.hidden_layers_list = { 2, num_decoder_layers / 2, num_decoder_layers - 3 };
         }
-        OPENVINO_ASSERT(eagle_rt_info.hidden_layers_list.size() == 3, "only exact 3 layer extraction are expected in eagle3");
+        OPENVINO_ASSERT(eagle_rt_info.hidden_layers_list.size() == 3, "only exactly 3 layer extraction are expected in eagle3");
     }
     return eagle_rt_info;
 }
@@ -91,7 +91,7 @@ ContinuousBatchingPipeline::ContinuousBatchingPipeline( const std::filesystem::p
     auto properties_without_draft_model = properties;
     auto draft_model_desr = utils::extract_draft_model_from_config(properties_without_draft_model);
     auto is_prompt_lookup_enabled = extract_prompt_lookup_from_config(properties_without_draft_model);
-    auto eagle_rt_info = extract_eagle_mode_from_config(draft_model_desr.properties, models_path);
+    auto eagle_rt_info = extract_eagle3_mode_from_config(draft_model_desr.properties, models_path);
 
     auto model = utils::read_model(models_path, properties);
     auto [properties_without_draft_model_without_gguf, enable_save_ov_model] = utils::extract_gguf_properties(properties_without_draft_model);
@@ -138,7 +138,7 @@ ContinuousBatchingPipeline::ContinuousBatchingPipeline(
     auto properties_without_draft_model = properties;
     auto draft_model_desr = utils::extract_draft_model_from_config(properties_without_draft_model);
     auto is_prompt_lookup_enabled = extract_prompt_lookup_from_config(properties_without_draft_model);
-    auto eagle_rt_info = extract_eagle_mode_from_config(draft_model_desr.properties, models_path);
+    auto eagle_rt_info = extract_eagle3_mode_from_config(draft_model_desr.properties, models_path);
     auto model = utils::read_model(models_path, properties_without_draft_model);
     auto [properties_without_draft_model_without_gguf, enable_save_ov_model] = utils::extract_gguf_properties(properties_without_draft_model);
     properties_without_draft_model_without_gguf[ov::cache_model_path.name()] = models_path;
@@ -188,7 +188,7 @@ ContinuousBatchingPipeline::ContinuousBatchingPipeline(
     auto properties_without_draft_model = properties;
     auto draft_model_desr = utils::extract_draft_model_from_config(properties_without_draft_model);
     auto is_prompt_lookup_enabled = extract_prompt_lookup_from_config(properties_without_draft_model);
-    auto eagle_rt_info = extract_eagle_mode_from_config(draft_model_desr.properties, std::filesystem::path(model_str));
+    auto eagle_rt_info = extract_eagle3_mode_from_config(draft_model_desr.properties, std::filesystem::path(model_str));
     auto model = utils::singleton_core().read_model(model_str, weights_tensor);
 
     auto rt_info = model->get_rt_info();

src/cpp/src/speculative_decoding/continuous_batching_for_speculative_decoding_impl.cpp

@@ -318,6 +318,7 @@ ContinuousBatchingPipeline::ContinuousBatchingForSpeculativeDecodingImpl::update
         size_t generated_len = request->get_context_len() >= request->get_prompt_len() ? request->get_context_len() - request->get_prompt_len() + 1 : 0;
         if (validate_length > 0) {
             if (generated_len > 0) {
+                // in eagle3 speculative mode, to rewind the processed tokens num to the stable kv position
                 request->update_processed_tokens_num(num_processed_tokens - result.removed_tokens_cnt + 1 - (validate_length - 1));
             }
         } else { // fast draft or main model for eagle speculative

tests/python_tests/test_continuous_batching.py

@@ -573,7 +573,7 @@ def run_extended_perf_metrics_collection(
         model_path = download_and_convert_model(model_id).models_path
         draft_model_path = None
         if draft_model_id is not None:
-            _,_, draft_model_path = download_and_convert_model(draft_model_id)
+            draft_model_path = download_and_convert_model(draft_model_id)
         ov_pipe = create_ov_pipeline(model_path, pipeline_type=pipeline_type, draft_model_path = draft_model_path)
         return ov_pipe.generate([prompt], generation_config).extended_perf_metrics
 
@@ -585,7 +585,7 @@ def run_extended_perf_metrics_collection(
         ignore_eos=True, 
         num_assistant_tokens=5,
     )
-    if  draft_model_id is None:
+    if draft_model_id is None:
         extended_perf_metrics = run_extended_perf_metrics_collection(
             main_model_id, generation_config, prompt, pipeline_type, draft_model_id
         )