Skip to content

Commit ee5aa1e

Browse files
authored
[ cb ] Remove penalties from Greedy config (openvinotoolkit#896)
1 parent cf88a7e commit ee5aa1e

File tree

2 files changed

+10
-13
lines changed

2 files changed

+10
-13
lines changed

samples/cpp/continuous_batching_benchmark/continuous_batching_benchmark.cpp

+10-7
Original file line numberDiff line numberDiff line change
@@ -121,6 +121,7 @@ Dataset filtered_dataset(const std::string& models_path, const std::string& data
121121

122122
ov::genai::GenerationConfig greedy_search = ov::genai::greedy();
123123
greedy_search.max_new_tokens = std::min(max_output_len, output_len);
124+
greedy_search.ignore_eos = true;
124125
greedy_search.repetition_penalty = 1.0;
125126
greedy_search.frequency_penalty = 0.0;
126127
greedy_search.presence_penalty = 0.0;
@@ -228,14 +229,16 @@ class GenerationInfo {
228229

229230
GenerationMetrics get_metrics() {
230231
GenerationMetrics generation_metrics;
231-
for (auto& sequenceInfoPair : sequences_info) {
232-
generation_metrics.mean_ttft += sequenceInfoPair.second.ttft;
233-
generation_metrics.mean_tpot += sequenceInfoPair.second.mean_tpot;
234-
generation_metrics.num_output_tokens += sequenceInfoPair.second.num_output_tokens;
232+
if (!sequences_info.empty()) {
233+
for (auto& sequenceInfoPair : sequences_info) {
234+
generation_metrics.mean_ttft += sequenceInfoPair.second.ttft;
235+
generation_metrics.mean_tpot += sequenceInfoPair.second.mean_tpot;
236+
generation_metrics.num_output_tokens += sequenceInfoPair.second.num_output_tokens;
237+
}
238+
generation_metrics.mean_ttft /= sequences_info.size();
239+
generation_metrics.mean_tpot /= sequences_info.size();
240+
generation_metrics.num_input_tokens = input_len;
235241
}
236-
generation_metrics.mean_ttft /= sequences_info.size();
237-
generation_metrics.mean_tpot /= sequences_info.size();
238-
generation_metrics.num_input_tokens = input_len;
239242
return generation_metrics;
240243
}
241244
};

src/cpp/src/generation_config.cpp

-6
Original file line numberDiff line numberDiff line change
@@ -172,12 +172,6 @@ GenerationConfig beam_search() {
172172

173173
GenerationConfig greedy() {
174174
GenerationConfig greedy_config;
175-
greedy_config.temperature = 0.0f;
176-
greedy_config.ignore_eos = true;
177-
greedy_config.num_return_sequences = 1;
178-
greedy_config.repetition_penalty = 3.0f;
179-
greedy_config.presence_penalty = 0.1f;
180-
greedy_config.frequency_penalty = 0.01f;
181175
greedy_config.max_new_tokens = 30;
182176
return greedy_config;
183177
}

0 commit comments

Comments
 (0)