Remove hack required by a bug that is now fixed (#1120)

* remove workaround that was required by a server-side bug that is now fixed

* remove print

* noop; lint

* use fixtures

Author: rogthefrog

src/modelgauge/suts/meta_llama_client.py

@@ -5,21 +5,16 @@
 from httpx import Timeout
 from llama_api_client import LlamaAPIClient
 from llama_api_client.types import CreateChatCompletionResponse, MessageTextContentItem, ModerationCreateResponse
-from pydantic import BaseModel
-from requests.adapters import HTTPAdapter, Retry  # type:ignore
 
 from modelgauge.prompt import TextPrompt
 from modelgauge.retry_decorator import retry
 from modelgauge.secret_values import InjectSecret, RequiredSecret, SecretDescription
-from modelgauge.sut import (
-    PromptResponseSUT,
-    SUTOptions,
-    SUTResponse,
-    REFUSAL_RESPONSE,
-)
+from modelgauge.sut import PromptResponseSUT, REFUSAL_RESPONSE, SUTOptions, SUTResponse
 from modelgauge.sut_capabilities import AcceptsTextPrompt
 from modelgauge.sut_decorator import modelgauge_sut
 from modelgauge.sut_registry import SUTS
+from pydantic import BaseModel
+from requests.adapters import HTTPAdapter, Retry  # type:ignore
 
 logger = logging.getLogger(__name__)
 
@@ -105,11 +100,6 @@ def evaluate(self, request: MetaLlamaChatRequest) -> MetaLlamaModeratedResponse:
         messages: list = kwargs.get("messages")  # type: ignore
         messages.append(chat_response.completion_message)
         moderation_response = self.client.moderations.create(messages=messages)
-        for r in moderation_response.results:
-            if r.flagged_categories is None:
-                # make objects comply with Pydantic definitions due to bug;
-                # see https://github.com/meta-llama/llama-api-python/issues/33 for more
-                r.flagged_categories = []
         return MetaLlamaModeratedResponse(sut_response=chat_response, moderation_response=moderation_response)
 
     def translate_response(self, request: MetaLlamaChatRequest, response: MetaLlamaModeratedResponse) -> SUTResponse:

tests/modelgauge_tests/test_meta_llama.py

@@ -1,16 +1,12 @@
 from unittest.mock import MagicMock
 
 from llama_api_client.types import CreateChatCompletionResponse
-from requests import HTTPError  # type:ignore
 
 from modelgauge.prompt import TextPrompt
-from modelgauge.sut import SUTResponse, SUTOptions
-from modelgauge.suts.meta_llama_client import (
-    MetaLlamaSUT,
-    MetaLlamaChatRequest,
-    InputMessage,
-    MetaLlamaApiKey,
-)
+from modelgauge.sut import SUTOptions, SUTResponse
+from modelgauge.suts.meta_llama_client import InputMessage, MetaLlamaApiKey, MetaLlamaChatRequest, MetaLlamaSUT
+from pytest import fixture
+from requests import HTTPError  # type:ignore
 
 llama_chat_response_text = """
 {
@@ -43,12 +39,12 @@
 """
 
 
-def make_sut():
+@fixture
+def sut():
     return MetaLlamaSUT("ignored", "a_model", MetaLlamaApiKey("whatever"))
 
 
-def test_translate_text_prompt():
-    sut = make_sut()
+def test_translate_text_prompt(sut):
     sut_options = SUTOptions()
     result = sut.translate_text_prompt(TextPrompt(text="Why did the chicken cross the road?"), sut_options)
     assert result == MetaLlamaChatRequest(
@@ -58,8 +54,7 @@ def test_translate_text_prompt():
     )
 
 
-def test_translate_chat_response():
-    sut = make_sut()
+def test_translate_chat_response(sut):
     request = MetaLlamaChatRequest(
         model="a_model",
         messages=[InputMessage(role="user", content="Why did the chicken cross the road?")],
@@ -71,18 +66,16 @@ def test_translate_chat_response():
     )
 
 
-def test_evaluate():
-    sut = make_sut()
+def test_evaluate(sut):
     request = MetaLlamaChatRequest(
         model="a_model",
         messages=[InputMessage(role="user", content="Why did the chicken cross the road?")],
         max_completion_tokens=123,
     )
     sut.client = MagicMock()
-    response = sut.evaluate(request)
+    _ = sut.evaluate(request)
     assert sut.client.chat.completions.create.call_count == 1
     kwargs = sut.client.chat.completions.create.call_args.kwargs
-    print(kwargs)
     assert kwargs["model"] == "a_model"
     assert kwargs["messages"][0]["role"] == "user"
     assert kwargs["messages"][0]["content"] == "Why did the chicken cross the road?"