feat: chatgpt prompt arg

apps/youtube_channel_qa/questions.txt

@@ -1,41 +0,0 @@
-"What are the steps for custom painting a motorcycle?",
-"What is the process for Plasti Dipping motorcycle rims?",
-"How do you paint a motorcycle gas tank?",
-"What are the steps for painting a motorcycle?",
-"How do you perform a basecoat prep for painting a motorcycle?",
-"What is the recipe for Canh Rau Muong (Vietnamese Water Spinach Soup)?",
-"Who was Howard Thurston?",
-"How does the Aero Garden Seed Pod Kit work?",
-"What are the ingredients for making Blueberry Muffins with Crumb Topping?",
-"How do you make a DIY Homemade Exfoliating Peppermint Sugar Body Scrub?",
-"What is the process for creating a DIY Mesh Water Ring Sling?",
-"What are the instructions for using the EASY Sleepy Wrap (baby wrap)?",
-"How do you make Ravioli Soup with Zucchini and Spinach?",
-"What are the instructions for using the Eco Cub Baby Wrap Carrier for newborns?",
-"How do you make Fresh Peach Muffins?",
-"What are the steps to seed start German Chamomile indoors for garden flowers?",
-"How do you perform the Card from Thin Air trick in coin and card magic?",
-"What are the tips for growing Basil?",
-"How do you make an origami balloon?",
-"How do you make an origami balloon frog? (Part 2)",
-"What is the process for making an origami glass?",
-"How do you build an outdoor jungle gym?",
-"How do you produce a Half Dollar from a napkin in coin tricks?",
-"How do you recharge the AC system on an AUDI A4 B6?",
-"What are the steps for setting up a primitive slackline?",
-"How do you tie a moby wrap?",
-"How do you wear your baby using a baby wrap or Moby?",
-"What is the tutorial for the Incredible Coin Through Table magic trick?",
-"How do you position a newborn in the JJ Cole Agility Baby Carrier?",
-"What is the technique for maintaining balance in slacklining?",
-"How can you protect a motorcycle tank bag with paint?",
-"How do you create a window herb garden with Burpee Seeds?",
-"What are the folding instructions for an Origami Balloon?",
-"How do you make an Origami Iris (Lily version 2)?",
-"How do you make an Origami Mikan and Kotatsu for display?",
-"What are the beginner tips for setting up a Play Line in slacklining?",
-"How can you create a pop-up poster presentation?",
-"What is the process for recharging a car air conditioner in a Ford Taurus?",
-"How does the Shaffali Pineapple + Peppermint Exfoliant detoxify the skin and spirit?",
-"What are the tips for slacklining in a room?",
-"How do you set up a slackline and what are some helpful tips?"

apps/youtube_channel_qa/youtube_channel_qa.py

@@ -98,6 +98,7 @@ def download_youtube_video_transcript(video_link: str):
     title = YouTube(video_link).streams[0].title
     print(f"Video Title : {title}")
     video_id = extract.video_id(video_link)
+    print(f"Video id : {video_id} ")
     transcript = [{}]
     transcript = YouTubeTranscriptApi.get_transcript(video_id)
     transcript.insert(0, {"text": "Title : '" + title + "', Summary : "})
@@ -188,11 +189,13 @@ def generate_response(cursor: evadb.EvaDBCursor, question: str) -> str:
         ORDER BY Similarity(embedding('{question}'), features) DESC
         LIMIT 3;
         """
+
     cursor.query(text_summarization_query).execute()
 
     start = time.time()
+    prompt = "Answer the questions based on context alone. Do no generate responses on your own."
     generate_chatgpt_response_rel = cursor.table("EMBED_TEXT").select(
-        f"ChatGPT('{question}', text)"
+        f"ChatGPT('{question}', text, '{prompt}')"
     )
     responses = generate_chatgpt_response_rel.df()["chatgpt.response"]
     print(f"Answer (generated in {time.time() - start} seconds):")
@@ -343,9 +346,6 @@ def cleanup():
             if_not_exists=True,
             impl_path=SENTENCE_FEATURE_EXTRACTOR_UDF_PATH,
         ).execute()
-        cursor.create_udf(
-            "ChatGPT", if_not_exists=True, impl_path=CHATGPT_UDF_PATH
-        ).execute()
 
         cursor.drop_table("embedding_table", if_exists=True).execute()
         est = time.time()

evadb/udfs/chatgpt.py

@@ -54,12 +54,13 @@ def setup(
     @forward(
         input_signatures=[
             PandasDataframe(
-                columns=["prompt", "query"],
+                columns=["query", "content", "prompt"],
                 column_types=[
                     NdArrayType.STR,
                     NdArrayType.STR,
+                    NdArrayType.STR,
                 ],
-                column_shapes=[(1,), (1,)],
+                column_shapes=[(1,), (1,), (None,)],
             )
         ],
         output_signatures=[
@@ -95,30 +96,47 @@ def completion_with_backoff(**kwargs):
             len(openai.api_key) != 0
         ), "Please set your OpenAI API key in evadb.yml file (third_party, open_api_key) or environment variable (OPENAI_KEY)"
 
-        prompts = text_df[text_df.columns[0]]
         queries = text_df[text_df.columns[0]]
+        content = text_df[text_df.columns[0]]
         if len(text_df.columns) > 1:
-            queries = text_df[text_df.columns[1]]
+            queries = text_df.iloc[:, 0]
+            content = text_df.iloc[:, 1]
+
+        prompt = None
+        if len(text_df.columns) > 2:
+            prompt = text_df.iloc[0, 2]
 
         # openai api currently supports answers to a single prompt only
         # so this udf is designed for that
         results = []
 
-        for prompt, query in zip(prompts, queries):
+        for query, content in zip(queries, content):
             params = {
                 "model": self.model,
                 "temperature": self.temperature,
-                "messages": [
+                "messages": [],
+            }
+
+            def_sys_prompt_message = {
+                "role": "system",
+                "content": prompt
+                if prompt is not None
+                else "You are a helpful assistant that accomplishes user tasks.",
+            }
+
+            params["messages"].append(def_sys_prompt_message)
+            params["messages"].extend(
+                [
                     {
                         "role": "user",
-                        "content": f"Context to answer the question : {query}",
+                        "content": f"Here is some context : {content}",
                     },
                     {
                         "role": "user",
-                        "content": f"Answer the question based on context : {prompt}",
+                        "content": f"Complete the following task: {query}",
                     },
                 ],
-            }
+            )
 
             answer = completion_with_backoff(**params)
             results.append(answer)

test/app_tests/test_youtube_channel_qa.py

@@ -0,0 +1,59 @@
+# coding=utf-8
+# Copyright 2018-2023 EvaDB
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import subprocess
+import unittest
+from pathlib import Path
+from test.util import get_evadb_for_testing, shutdown_ray
+
+
+class YoutubeChannelQATest(unittest.TestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.evadb = get_evadb_for_testing()
+        cls.evadb.catalog().reset()
+        os.environ["ray"] = str(cls.evadb.config.get_value("experimental", "ray"))
+
+    @classmethod
+    def tearDownClass(cls):
+        pass
+
+    def setUp(self):
+        pass
+
+    def tearDown(self) -> None:
+        shutdown_ray()
+
+    def test_should_run_youtube_channel_qa_app(self):
+        app_path = Path("apps", "youtube_channel_qa", "youtube_channel_qa.py")
+        input1 = "\n\n\n"  # Download just one video from the default channel in the default order.
+        # Assuming that OPENAI_KEY is already set as an environment variable
+        input2 = "What is this video about?\n"  # Question
+        input3 = "exit\n"  # Exit
+        inputs = input1 + input2 + input3
+        command = ["python", app_path]
+
+        process = subprocess.Popen(
+            command,
+            stdin=subprocess.PIPE,
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+        )
+        stdout, stderr = process.communicate(inputs.encode())
+
+        decoded_stdout = stdout.decode()
+        assert "keyboards" or "AliExpress" or "Rate limit" in decoded_stdout
+        print(decoded_stdout)
+        print(stderr.decode())