Dev (#2940)

# Purpose

Please provide a high-level overview of what this pull request aims to
achieve.

# Changes Made

Please provide a detailed list of the changes made in this pull request.

1.
2.
3.

# Additional Notes

Please provide any additional notes or screenshots here.
When you make a PR, please ping us on Discord at
http://discord.gg/sweep.

---------

Co-authored-by: Kevin Lu <[email protected]>

sweepai/agents/assistant_wrapper.py

@@ -42,7 +42,7 @@ def openai_retry_with_timeout(call, *args, num_retries=3, timeout=5, **kwargs):
         try:
             return call(*args, **kwargs, timeout=timeout)
         except Exception as e:
-            logger.error(f"Retry {attempt + 1} failed with error: {e}")
+            logger.exception(f"Retry {attempt + 1} failed with error: {e}")
             error_message = str(e)
     raise Exception(
         f"Maximum retries reached. The call failed for call {error_message}"

sweepai/api.py

@@ -967,6 +967,8 @@ def worker_wrapper():
 
         thread = threading.Thread(target=worker_wrapper)
         thread.start()
+        thread_killer = threading.Thread(target=delayed_kill, args=(thread,))
+        thread_killer.start()
         return {"success": True}
 
 

sweepai/core/chat.py

@@ -63,7 +63,7 @@ class ChatGPT(BaseModel):
         )
     ]
     prev_message_states: list[list[Message]] = []
-    model: ChatModel = "gpt-4-0613"
+    model: ChatModel = "gpt-4-1106-preview"
     chat_logger: ChatLogger | None
     human_message: HumanMessagePrompt | None = None
     file_change_paths: list[str] = []
@@ -368,11 +368,6 @@ async def acall_openai(
             messages_dicts.append(message_dict)
 
         gpt_4_buffer = 800
-        # if int(messages_length) + gpt_4_buffer < 6000 and model == DEFAULT_GPT4_32K_MODEL:
-        #     model = "gpt-4-0613"
-        #     max_tokens = (
-        #         model_to_max_tokens[model] - int(messages_length) - gpt_4_buffer
-        #     )  # this is for the function tokens
         if "gpt-4" in model:
             max_tokens = min(max_tokens, 4096)
         # Fix for self hosting where TPM limit is super low for GPT-4

sweepai/core/context_pruning.py

@@ -1,12 +1,14 @@
 import json
 import re
+import textwrap
 import time
 
 from attr import dataclass
 from loguru import logger
 from openai.types.beta.thread import Thread
 from openai.types.beta.threads.run import Run
 
+from sweepai.agents.assistant_function_modify import MAX_CHARS
 from sweepai.agents.assistant_wrapper import client, openai_retry_with_timeout
 from sweepai.config.server import IS_SELF_HOSTED
 from sweepai.core.entities import Snippet
@@ -275,6 +277,7 @@ def get_relevant_context(
             unformatted_user_prompt=unformatted_user_prompt,
             query=query,
         )
+        messages = textwrap.wrap(user_prompt, MAX_CHARS)
         assistant = openai_retry_with_timeout(
             client.beta.assistants.create,
             name="Relevant Files Assistant",
@@ -283,12 +286,13 @@ def get_relevant_context(
             model=model,
         )
         thread = openai_retry_with_timeout(client.beta.threads.create)
-        _ = openai_retry_with_timeout(
-            client.beta.threads.messages.create,
-            thread.id,
-            role="user",
-            content=f"{user_prompt}",
-        )
+        for content in messages:
+            _ = openai_retry_with_timeout(
+                client.beta.threads.messages.create,
+                thread.id,
+                role="user",
+                content=content,
+            )
         run = openai_retry_with_timeout(
             client.beta.threads.runs.create,
             thread_id=thread.id,

sweepai/handlers/on_ticket.py

@@ -356,19 +356,15 @@ def delete_old_prs():
                 direction="desc",
                 base=SweepConfig.get_branch(repo),
             )
-            checked_pr_count = 0
-            for pr in tqdm(prs):
+            for pr in tqdm(prs.get_page(0)):
                 # # Check if this issue is mentioned in the PR, and pr is owned by bot
                 # # This is done in create_pr, (pr_description = ...)
-                if checked_pr_count >= 40:
-                    break
                 if (
                     pr.user.login == CURRENT_USERNAME
                     and f"Fixes #{issue_number}.\n" in pr.body
                 ):
                     success = safe_delete_sweep_branch(pr, repo)
                     break
-                checked_pr_count += 1
 
         fire_and_forget_wrapper(delete_old_prs)()
 

sweepai/utils/github_utils.py

@@ -44,7 +44,7 @@ def get_jwt():
 
 
 def get_token(installation_id: int):
-    if installation_id < 0:
+    if int(installation_id) < 0:
         return os.environ["GITHUB_PAT"]
     for timeout in [5.5, 5.5, 10.5]:
         try:

sweepai/utils/openai_proxy.py

@@ -114,6 +114,7 @@ def determine_openai_engine(self, model):
         elif (
             model == "gpt-4"
             or model == "gpt-4-0613"
+            or model == "gpt-4-1106-preview"
             and OPENAI_API_ENGINE_GPT4 is not None
         ):
             engine = OPENAI_API_ENGINE_GPT4

sweepai/utils/ticket_utils.py

@@ -1,5 +1,4 @@
 import traceback
-from threading import Thread
 from time import time
 
 from loguru import logger
@@ -244,13 +243,14 @@ def fire_and_forget_wrapper(call):
     """
 
     def wrapper(*args, **kwargs):
-        def run_in_thread(call, *a, **kw):
-            try:
-                call(*a, **kw)
-            except:
-                pass
+        return call(*args, **kwargs)
+        # def run_in_thread(call, *a, **kw):
+        #     try:
+        #         call(*a, **kw)
+        #     except:
+        #         pass
 
-        thread = Thread(target=run_in_thread, args=(call,) + args, kwargs=kwargs)
-        thread.start()
+        # thread = Thread(target=run_in_thread, args=(call,) + args, kwargs=kwargs)
+        # thread.start()
 
     return wrapper