Merge pull request #237 from Kav-K/request-testing

Request testing

.gitignore

@@ -11,4 +11,5 @@ usage.txt
 /dalleimages
 /indexes
 /audiotemp
-/pickles
+/pickles
+.idea

gpt3discord.py

@@ -33,7 +33,7 @@ from services.environment_service import EnvService
 from models.openai_model import Model
 
 
-__version__ = "11.1.4"
+__version__ = "11.1.5"
 
 
 PID_FILE = Path("bot.pid")

models/openai_model.py

@@ -633,10 +633,14 @@ class Model:
             else:
                 await self.usage_service.update_usage(tokens_used)
         except Exception as e:
-            raise ValueError(
+            traceback.print_exc()
-                "The API returned an invalid response: "
+            if 'error' in response:
-                + str(response["error"]["message"])
+                raise ValueError(
-            ) from e
+                    "The API returned an invalid response: "
+                    + str(response["error"]["message"])
+                ) from e
+            else:
+                raise ValueError("The API returned an invalid response") from e
 
     @backoff.on_exception(
         backoff.expo,
@@ -1010,13 +1014,11 @@ class Model:
         stop=None,
         custom_api_key=None,
         is_chatgpt_request=False,
-    ) -> (
-        Tuple[dict, bool]
     ):  # The response, and a boolean indicating whether or not the context limit was reached.
         # Validate that  all the parameters are in a good state before we send the request
 
         if not max_tokens_override:
-            if model:
+            if model and model not in Models.GPT4_MODELS and model not in Models.CHATGPT_MODELS:
                 max_tokens_override = Models.get_max_tokens(model) - tokens
 
         print(f"The prompt about to be sent is {prompt}")
@@ -1057,7 +1059,7 @@ class Model:
                     headers=headers,
                 ) as resp:
                     response = await resp.json()
-                    # print(f"Payload -> {payload}")
+                    print(f"Payload -> {payload}")
                     # Parse the total tokens used for this request and response pair from the response
                     await self.valid_text_request(
                         response, model=self.model if model is None else model

pyproject.toml

@@ -41,7 +41,9 @@ dependencies = [
 "langchain==0.0.115",
 "unidecode==1.3.6",
 "tqdm==4.64.1",
-"docx2txt==0.8"
+"docx2txt==0.8",
+"pytest-asyncio==0.21.0",
+"pytest~=7.2.2"
 ]
 
 dynamic = ["version"]

requirements.txt

@@ -24,3 +24,5 @@ openai-whisper
 unidecode==1.3.6
 tqdm==4.64.1
 docx2txt==0.8
+pytest-asyncio==0.21.0
+pytest~=7.2.2

requirements_base.txt

@@ -21,4 +21,6 @@ python-pptx==0.6.21
 langchain==0.0.115
 unidecode==1.3.6
 tqdm==4.64.1
-docx2txt==0.8
+docx2txt==0.8
+pytest-asyncio==0.21.0
+pytest~=7.2.2

tests/test_requests.py

@@ -0,0 +1,51 @@
+from pathlib import Path
+
+import pytest
+from models.openai_model import Model
+from transformers import GPT2TokenizerFast
+
+from services.usage_service import UsageService
+
+
+# Non-ChatGPT -> TODO: make generic test and loop through text models
+@pytest.mark.asyncio
+async def test_send_req():
+
+    usage_service = UsageService(Path("../tests"))
+    model = Model(usage_service)
+    prompt = 'how many hours are in a day?'
+    tokens = len(GPT2TokenizerFast.from_pretrained("gpt2")(prompt)["input_ids"])
+    res = await model.send_request(prompt, tokens)
+    assert '24' in res['choices'][0]['text']
+
+
+# ChatGPT version
+@pytest.mark.asyncio
+async def test_send_req_gpt():
+
+    usage_service = UsageService(Path("../tests"))
+    model = Model(usage_service)
+    prompt = 'how many hours are in a day?'
+    res = await model.send_request(prompt, None, is_chatgpt_request=True, model="gpt-3.5-turbo")
+    assert '24' in res['choices'][0]['message']['content']
+
+
+# GPT4 version
+@pytest.mark.asyncio
+async def test_send_req_gpt4():
+    usage_service = UsageService(Path("../tests"))
+    model = Model(usage_service)
+    prompt = 'how many hours are in a day?'
+    res = await model.send_request(prompt, None, is_chatgpt_request=True, model="gpt-4")
+    assert '24' in res['choices'][0]['message']['content']
+
+
+# Edit request -> currently broken due to endpoint
+# @pytest.mark.asyncio
+# async def test_send_edit_req():
+#     usage_service = UsageService(Path("../tests"))
+#     model = Model(usage_service)
+#     text = 'how many hours are in a day?'
+#     res = await model.send_edit_request(text, codex=True)
+#     assert '24' in res['choices'][0]['text']
+