Skip to content

Commit

Permalink
fix full_response false and no mapping exceptions (opensearch-project…
Browse files Browse the repository at this point in the history 
…#2944)

Signed-off-by: Mingshi Liu <[email protected]>
  • Loading branch information
@mingshl
mingshl committed Sep 13, 2024
1 parent bbaea1c commit 853efd6
Show file tree
Hide file tree
Showing 2 changed files with 194 additions and 21 deletions.
2 changes: 1 addition & 1 deletion plugin/src/main/java/org/opensearch/ml/processor/MLInferenceSearchResponseProcessor.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -634,7 +634,7 @@ private static Map<String, String> getDefaultOutputMapping(Integer mappingIndex,
Map<String, String> outputMapping;
if (processOutputMap == null || processOutputMap.size() == 0) {
outputMapping = new HashMap<>();
outputMapping.put(DEFAULT_OUTPUT_FIELD_NAME, "$." + DEFAULT_OUTPUT_FIELD_NAME);
outputMapping.put(DEFAULT_OUTPUT_FIELD_NAME, null);
} else {
outputMapping = processOutputMap.get(mappingIndex);
}
Expand Down
213 changes: 193 additions & 20 deletions ...in/src/test/java/org/opensearch/ml/processor/MLInferenceSearchResponseProcessorTests.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -351,6 +351,158 @@ public void onFailure(Exception e) {
verify(client, times(1)).execute(any(), any(), any());
}

/**
* Tests create processor with one_to_one is false
* with custom prompt
* with many to one prediction, 5 documents in hits are calling 1 prediction tasks
* with full response path false and no output mapping is provided
* @throws Exception if an error occurs during the test
*/
public void testProcessResponseManyToOneWithCustomPromptFullResponsePathFalse() throws Exception {

String documentField = "text";
String modelInputField = "context";
List<Map<String, String>> inputMap = new ArrayList<>();
Map<String, String> input = new HashMap<>();
input.put(modelInputField, documentField);
inputMap.add(input);

Map<String, String> modelConfig = new HashMap<>();
modelConfig
.put(
"prompt",
"\\n\\nHuman: You are a professional data analyst. You will always answer question based on the given context first. If the answer is not directly shown in the context, you will analyze the data and find the answer. If you don't know the answer, just say I don't know. Context: ${parameters.context}. \\n\\n Human: please summarize the documents \\n\\n Assistant:"
);
MLInferenceSearchResponseProcessor responseProcessor = new MLInferenceSearchResponseProcessor(
"model1",
inputMap,
null,
modelConfig,
DEFAULT_MAX_PREDICTION_TASKS,
PROCESSOR_TAG,
DESCRIPTION,
false,
"remote",
false,
false,
false,
"{ \"parameters\": ${ml_inference.parameters} }",
client,
TEST_XCONTENT_REGISTRY_FOR_QUERY,
false
);

SearchRequest request = getSearchRequest();
String fieldName = "text";
SearchResponse response = getSearchResponse(5, true, fieldName);
Map<String, Object> predictionResult = ImmutableMap.of("response", "here is a summary of the documents");

ModelTensor modelTensor = ModelTensor.builder().dataAsMap(predictionResult).build();
ModelTensors modelTensors = ModelTensors.builder().mlModelTensors(Arrays.asList(modelTensor)).build();
ModelTensorOutput mlModelTensorOutput = ModelTensorOutput.builder().mlModelOutputs(Arrays.asList(modelTensors)).build();

doAnswer(invocation -> {
ActionListener<MLTaskResponse> actionListener = invocation.getArgument(2);
actionListener.onResponse(MLTaskResponse.builder().output(mlModelTensorOutput).build());
return null;
}).when(client).execute(any(), any(), any());

ActionListener<SearchResponse> listener = new ActionListener<>() {
@Override
public void onResponse(SearchResponse newSearchResponse) {
assertEquals(newSearchResponse.getHits().getHits().length, 5);
assertEquals(newSearchResponse.getHits().getHits()[0].getSourceAsMap().get("inference_results"), predictionResult);
assertEquals(newSearchResponse.getHits().getHits()[1].getSourceAsMap().get("inference_results"), predictionResult);
assertEquals(newSearchResponse.getHits().getHits()[2].getSourceAsMap().get("inference_results"), predictionResult);
assertEquals(newSearchResponse.getHits().getHits()[3].getSourceAsMap().get("inference_results"), predictionResult);
}

@Override
public void onFailure(Exception e) {
throw new RuntimeException(e);
}

};
responseProcessor.processResponseAsync(request, response, responseContext, listener);
verify(client, times(1)).execute(any(), any(), any());
}

/**
* Tests create processor with one_to_one is false
* with custom prompt
* with many to one prediction, 5 documents in hits are calling 1 prediction tasks
* with full response path true and no output mapping is provided
* @throws Exception if an error occurs during the test
*/
public void testProcessResponseManyToOneWithCustomPromptFullResponsePathTrue() throws Exception {

String documentField = "text";
String modelInputField = "context";
List<Map<String, String>> inputMap = new ArrayList<>();
Map<String, String> input = new HashMap<>();
input.put(modelInputField, documentField);
inputMap.add(input);

Map<String, String> modelConfig = new HashMap<>();
modelConfig
.put(
"prompt",
"\\n\\nHuman: You are a professional data analyst. You will always answer question based on the given context first. If the answer is not directly shown in the context, you will analyze the data and find the answer. If you don't know the answer, just say I don't know. Context: ${parameters.context}. \\n\\n Human: please summarize the documents \\n\\n Assistant:"
);
MLInferenceSearchResponseProcessor responseProcessor = new MLInferenceSearchResponseProcessor(
"model1",
inputMap,
null,
modelConfig,
DEFAULT_MAX_PREDICTION_TASKS,
PROCESSOR_TAG,
DESCRIPTION,
false,
"remote",
true,
false,
false,
"{ \"parameters\": ${ml_inference.parameters} }",
client,
TEST_XCONTENT_REGISTRY_FOR_QUERY,
false
);

SearchRequest request = getSearchRequest();
String fieldName = "text";
SearchResponse response = getSearchResponse(5, true, fieldName);
Map<String, Object> predictionResult = ImmutableMap.of("response", "here is a summary of the documents");
ModelTensor modelTensor = ModelTensor.builder().dataAsMap(predictionResult).build();
ModelTensors modelTensors = ModelTensors.builder().mlModelTensors(Arrays.asList(modelTensor)).build();
ModelTensorOutput mlModelTensorOutput = ModelTensorOutput.builder().mlModelOutputs(Arrays.asList(modelTensors)).build();
Map<String, Object> fullPredictionResult = generateInferenceResult("here is a summary of the documents");

doAnswer(invocation -> {
ActionListener<MLTaskResponse> actionListener = invocation.getArgument(2);
actionListener.onResponse(MLTaskResponse.builder().output(mlModelTensorOutput).build());
return null;
}).when(client).execute(any(), any(), any());

ActionListener<SearchResponse> listener = new ActionListener<>() {
@Override
public void onResponse(SearchResponse newSearchResponse) {
assertEquals(newSearchResponse.getHits().getHits().length, 5);
assertEquals(newSearchResponse.getHits().getHits()[0].getSourceAsMap().get("inference_results"), fullPredictionResult);
assertEquals(newSearchResponse.getHits().getHits()[1].getSourceAsMap().get("inference_results"), fullPredictionResult);
assertEquals(newSearchResponse.getHits().getHits()[2].getSourceAsMap().get("inference_results"), fullPredictionResult);
assertEquals(newSearchResponse.getHits().getHits()[3].getSourceAsMap().get("inference_results"), fullPredictionResult);
}

@Override
public void onFailure(Exception e) {
throw new RuntimeException(e);
}

};
responseProcessor.processResponseAsync(request, response, responseContext, listener);
verify(client, times(1)).execute(any(), any(), any());
}

/**
* Tests create processor with one_to_one is true
* with no mapping provided
Expand Down Expand Up @@ -401,23 +553,23 @@ public void onResponse(SearchResponse newSearchResponse) {
assertEquals(newSearchResponse.getHits().getHits().length, 5);
assertEquals(
newSearchResponse.getHits().getHits()[0].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
assertEquals(
newSearchResponse.getHits().getHits()[1].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
assertEquals(
newSearchResponse.getHits().getHits()[2].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
assertEquals(
newSearchResponse.getHits().getHits()[3].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
assertEquals(
newSearchResponse.getHits().getHits()[4].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
}

Expand Down Expand Up @@ -482,23 +634,23 @@ public void onResponse(SearchResponse newSearchResponse) {
assertEquals(newSearchResponse.getHits().getHits().length, 5);
assertEquals(
newSearchResponse.getHits().getHits()[0].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
assertEquals(
newSearchResponse.getHits().getHits()[1].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
assertEquals(
newSearchResponse.getHits().getHits()[2].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
assertEquals(
newSearchResponse.getHits().getHits()[3].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
assertEquals(
newSearchResponse.getHits().getHits()[4].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
}

Expand Down Expand Up @@ -1893,23 +2045,23 @@ public void onResponse(SearchResponse newSearchResponse) {
assertEquals(newSearchResponse.getHits().getHits().length, 5);
assertEquals(
newSearchResponse.getHits().getHits()[0].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
assertEquals(
newSearchResponse.getHits().getHits()[1].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
assertEquals(
newSearchResponse.getHits().getHits()[2].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
assertEquals(
newSearchResponse.getHits().getHits()[3].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
assertEquals(
newSearchResponse.getHits().getHits()[4].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
}

Expand Down Expand Up @@ -1973,23 +2125,23 @@ public void onResponse(SearchResponse newSearchResponse) {
assertEquals(newSearchResponse.getHits().getHits().length, 5);
assertEquals(
newSearchResponse.getHits().getHits()[0].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
assertEquals(
newSearchResponse.getHits().getHits()[1].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
assertEquals(
newSearchResponse.getHits().getHits()[2].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
assertEquals(
newSearchResponse.getHits().getHits()[3].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
assertEquals(
newSearchResponse.getHits().getHits()[4].getSourceAsMap().get(DEFAULT_OUTPUT_FIELD_NAME).toString(),
"[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]"
"{inference_results=[{output=[{dataAsMap={response=[0.0, 1.0, 2.0, 3.0, 4.0]}}]}]}"
);
}

Expand Down Expand Up @@ -3054,6 +3206,27 @@ private static SearchRequest getSearchRequest() {
return request;
}

private static Map<String, Object> generateInferenceResult(String response) {
Map<String, Object> inferenceResult = new HashMap<>();
List<Map<String, Object>> inferenceResults = new ArrayList<>();

Map<String, Object> outputMap = new HashMap<>();
List<Map<String, Object>> outputs = new ArrayList<>();

Map<String, Object> responseOutput = new HashMap<>();
Map<String, String> dataAsMap = new HashMap<>();
dataAsMap.put("response", response);
responseOutput.put("dataAsMap", dataAsMap);

outputs.add(responseOutput);
outputMap.put("output", outputs);

inferenceResults.add(outputMap);
inferenceResult.put("inference_results", inferenceResults);

return inferenceResult;
}

/**
* Helper method to create an instance of the MLInferenceSearchResponseProcessor with the specified parameters in
* single pair of input and output mapping.
Expand Down

0 comments on commit 853efd6

Please sign in to comment.