[ML] Refactor OpenAI to use ConstructingObjectParser and consolidate class locations #124380
Conversation
jonathan-buttner
added
>non-issue
:ml
| ensureExpectedToken(XContentParser.Token.START_OBJECT, token, jsonParser); | ||
|
|
||
| positionParserAtTokenAfterField(jsonParser, "choices", FAILED_TO_FIND_FIELD_TEMPLATE); | ||
| try (var p = XContentFactory.xContent(XContentType.JSON).createParser(XContentParserConfiguration.EMPTY, response.body())) { |
There was a problem hiding this comment.
The major changes here are to use the ConstructingObjectParser instead of iterating by token.
| @@ -0,0 +1,107 @@ | |||
| /* | |||
There was a problem hiding this comment.
This class was moved and transitioned to use the ConstructingObjectParser.
| @@ -1,110 +0,0 @@ | |||
| /* | |||
There was a problem hiding this comment.
This was moved to external.openai and transitioned to use the ConstructingObjectParser.
| @@ -219,7 +219,7 @@ public void testCreate_AzureOpenAiEmbeddingsModel_FailsFromInvalidResponseFormat | |||
| PlainActionFuture<InferenceServiceResults> listener = new PlainActionFuture<>(); | |||
| action.execute(new DocumentsOnlyInput(List.of("abc")), InferenceAction.Request.DEFAULT_TIMEOUT, listener); | |||
|
|
|||
| var failureCauseMessage = "Failed to find required field [data] in OpenAI embeddings response"; | |||
| var failureCauseMessage = "Required [data]"; | |||
There was a problem hiding this comment.
If we encounter a parsing failure where a field is missing, the error message will be decorated on the way back in one of the upstream listeners. The error message will have the openai information included elsewhere.
| @@ -533,7 +533,7 @@ public void testCreate_OpenAiChatCompletionModel_FailsFromInvalidResponseFormat( | |||
| PlainActionFuture<InferenceServiceResults> listener = new PlainActionFuture<>(); | |||
| action.execute(new ChatCompletionInput(List.of("abc")), InferenceAction.Request.DEFAULT_TIMEOUT, listener); | |||
|
|
|||
| var failureCauseMessage = "Failed to find required field [choices] in OpenAI chat completions response"; | |||
| var failureCauseMessage = "Required [choices]"; | |||
| var thrownException = expectThrows(ElasticsearchStatusException.class, () -> listener.actionGet(TIMEOUT)); | |||
| assertThat( | |||
| thrownException.getMessage(), | |||
There was a problem hiding this comment.
Here's an example of the message containing openai and chat completions information already.
|
Pinging @elastic/ml-core (Team:ML) |
This PR refactors the OpenAI response parsing logic.
external.openai