Merge remote-tracking branch 'origin/main' into mistral-ignore-bot

Author: dkalinowski

demos/common/export_models/export_model.py

@@ -659,7 +659,7 @@ def export_image_generation_model(model_repository_path, source_model, model_nam
     export_text_generation_model(args['model_repository_path'], args['source_model'], args['model_name'], args['precision'], template_parameters, args['config_file_path'])
 
 elif args['task'] == 'embeddings_ov':
-    export_embeddings_model_ov(args['model_repository_path'], args['source_model'], args['model_name'],  args['precision'], template_parameters)
+    export_embeddings_model_ov(args['model_repository_path'], args['source_model'], args['model_name'],  args['precision'], template_parameters, args['config_file_path'], args['truncate'])
 
 elif args['task'] == 'rerank':
     export_rerank_model(args['model_repository_path'], args['source_model'], args['model_name'] ,args['precision'], template_parameters, str(args['version']), args['config_file_path'], args['max_doc_length'])

src/test/llm/output_parsers/mistral_output_parser_test.cpp

@@ -160,29 +160,29 @@ TEST_F(MistralOutputParserTest, ParseToolCallOutputWithContentAndNoToolCalls) {
 }
 
 TEST_F(MistralOutputParserTest, ParseToolCallOutputWithContentAndSingleToolCall) {
-    std::string input = "This is a content part and next will be a tool call.\n\n [TOOL_CALLS][{\"name\": \"example_tool\", \"arguments\": {\"arg1\": \"value1\", \"arg2\": 42}}]</s>";
+    std::string input = "This is a content part and next will be a tool call.\n\n[TOOL_CALLS][{\"name\": \"example_tool\", \"arguments\": {\"arg1\": \"value1\", \"arg2\": 42}}]</s>";
     for (bool immediateParsing : {false, true}) {
         auto generatedTensor = mistralTokenizer->encode(input, ov::genai::add_special_tokens(false)).input_ids;
         std::vector<int64_t> generatedTokens(generatedTensor.data<int64_t>(), generatedTensor.data<int64_t>() + generatedTensor.get_size());
         ParsedOutput parsedOutput = immediateParsing ? outputParserWithImmediateToolParsing->parse(generatedTokens, true) : outputParserWithRegularToolParsing->parse(generatedTokens, true);
-        EXPECT_EQ(parsedOutput.content, "This is a content part and next will be a tool call.\n\n[{\"name\": \"example_tool\", \"arguments\": {\"arg1\": \"value1\", \"arg2\": 42}}]");
+        EXPECT_EQ(parsedOutput.content, "This is a content part and next will be a tool call.\n\n [{\"name\": \"example_tool\", \"arguments\": {\"arg1\": \"value1\", \"arg2\": 42}}]");
         EXPECT_EQ(parsedOutput.reasoning, "");
         ASSERT_EQ(parsedOutput.toolCalls.size(), 0);
     }
 }
 TEST_F(MistralOutputParserTest, ParseToolCallOutputWithContentOnBothSidesAndSingleToolCall) {
-    std::string input = "This is a content part and next will be a tool call.\n\n [TOOL_CALLS][{\"name\": \"example_tool\", \"arguments\": {\"arg1\": \"value1\", \"arg2\": 42}}]</s> This is a content part after tool call.";
+    std::string input = "This is a content part and next will be a tool call.\n\n[TOOL_CALLS][{\"name\": \"example_tool\", \"arguments\": {\"arg1\": \"value1\", \"arg2\": 42}}]</s> This is a content part after tool call.";
     for (bool immediateParsing : {false, true}) {
         auto generatedTensor = mistralTokenizer->encode(input, ov::genai::add_special_tokens(false)).input_ids;
         std::vector<int64_t> generatedTokens(generatedTensor.data<int64_t>(), generatedTensor.data<int64_t>() + generatedTensor.get_size());
         ParsedOutput parsedOutput = immediateParsing ? outputParserWithImmediateToolParsing->parse(generatedTokens, true) : outputParserWithRegularToolParsing->parse(generatedTokens, true);
-        EXPECT_EQ(parsedOutput.content, "This is a content part and next will be a tool call.\n\n[{\"name\": \"example_tool\", \"arguments\": {\"arg1\": \"value1\", \"arg2\": 42}}] This is a content part after tool call.");
+        EXPECT_EQ(parsedOutput.content, "This is a content part and next will be a tool call.\n\n [{\"name\": \"example_tool\", \"arguments\": {\"arg1\": \"value1\", \"arg2\": 42}}] This is a content part after tool call.");
         EXPECT_EQ(parsedOutput.reasoning, "");
         ASSERT_EQ(parsedOutput.toolCalls.size(), 0);
     }
 }
 TEST_F(MistralOutputParserTest, ParseToolCallOutputWithMultipleToolCallsReturnsContentOnly) {
-    std::string input = "[TOOL_CALLS][{\"name\": \"tool1\", \"arguments\": {\"a\": 1}}]</s> \n\nThis is some content\n\n [TOOL_CALLS][{\"name\": \"tool2\", \"arguments\": {\"b\": 2}}]</s>";
+    std::string input = "[TOOL_CALLS][{\"name\": \"tool1\", \"arguments\": {\"a\": 1}}]</s> \n\nThis is some content\n\n[TOOL_CALLS][{\"name\": \"tool2\", \"arguments\": {\"b\": 2}}]</s>";
     for (bool immediateParsing : {false, true}) {
         std::string testInput = input;
         if (immediateParsing) {
@@ -192,7 +192,7 @@ TEST_F(MistralOutputParserTest, ParseToolCallOutputWithMultipleToolCallsReturnsC
         std::vector<int64_t> generatedTokens(generatedTensor.data<int64_t>(), generatedTensor.data<int64_t>() + generatedTensor.get_size());
         ParsedOutput parsedOutput = immediateParsing ? outputParserWithImmediateToolParsing->parse(generatedTokens, true) : outputParserWithRegularToolParsing->parse(generatedTokens, true);
         // Same expected content as tokenizer does not add special tokens
-        EXPECT_EQ(parsedOutput.content, "[{\"name\": \"tool1\", \"arguments\": {\"a\": 1}}] \n\nThis is some content\n\n[{\"name\": \"tool2\", \"arguments\": {\"b\": 2}}]");
+        EXPECT_EQ(parsedOutput.content, "[{\"name\": \"tool1\", \"arguments\": {\"a\": 1}}] \n\nThis is some content\n\n [{\"name\": \"tool2\", \"arguments\": {\"b\": 2}}]");
         EXPECT_EQ(parsedOutput.reasoning, "");
         ASSERT_EQ(parsedOutput.toolCalls.size(), 0);
     }