hax to run against local starcoder/wizardcoder models

aider/coders/base_coder.py

@@ -737,7 +737,8 @@ def show_send_output_stream(self, completion, silent):
 
  for chunk in completion:
  if chunk.choices[0].finish_reason == "length":
- raise ExhaustedContextWindow()
+ pass
+ #raise ExhaustedContextWindow()
 
  try:
  func = chunk.choices[0].delta.function_call

aider/coders/wholefile_coder.py

@@ -71,6 +71,7 @@ def update_files(self, mode="update"):
  if i > 0:
  fname_source = "block"
  fname = lines[i - 1].strip()
+ fname = fname.strip("*") # handle **filename.py** from wizardcoder
  # Did gpt prepend a bogus dir? It especially likes to
  # include the path/to prefix from the one-shot example in
  # the prompt.

aider/coders/wholefile_prompts.py

@@ -6,12 +6,10 @@
 class WholeFilePrompts(CoderPrompts):
  main_system = """Act as an expert software developer.
 Take requests for changes to the supplied code.
-If the request is ambiguous, ask questions.
 
-Once you understand the request you MUST:
-1. Determine if any code changes are needed.
+1. Determine what code changes are needed.
 2. Explain any needed changes.
-3. If changes are needed, output a copy of each file that needs changes.
+3. Output a copy of each file that needs changes.
 """
 
  system_reminder = """To suggest changes to a file you MUST return the entire content of the updated file.

aider/commands.py

@@ -17,7 +17,7 @@ class Commands:
  def __init__(self, io, coder):
  self.io = io
  self.coder = coder
- self.tokenizer = tiktoken.encoding_for_model(coder.main_model.name)
+ self.tokenizer = tiktoken.encoding_for_model('gpt-4') #coder.main_model.name)
 
  def is_command(self, inp):
  if inp[0] == "/":

aider/models.py

@@ -28,7 +28,8 @@ def __init__(self, name):
  tokens = t
 
  if tokens is None:
- raise ValueError(f"Unknown context window size for model: {name}")
+ tokens = 1
+ #raise ValueError(f"Unknown context window size for model: {name}")
 
  self.max_context_tokens = tokens * 1024
 
@@ -46,7 +47,7 @@ def __init__(self, name):
 
  return
 
- if self.is_gpt35():
+ if self.is_gpt35() or True:
  self.edit_format = "whole"
  self.always_available = True
 

benchmark/Dockerfile

@@ -1,5 +1,5 @@
 FROM python:3.8-slim
-RUN apt-get update && apt-get install -y less git
+RUN apt-get update && apt-get install -y less git vim
 COPY requirements.txt /aider/requirements.txt
 RUN pip install lox typer pandas matplotlib imgcat aider-chat
 RUN pip install --upgrade pip && pip install -r /aider/requirements.txt

benchmark/benchmark.py

@@ -1,4 +1,4 @@
-#!/usr/bin/env python
+#!/usr/bin/env python3
 
 import datetime
 import json
@@ -509,6 +509,8 @@ def run_test(
  print("fnames:", show_fnames)
 
  openai.api_key = os.environ["OPENAI_API_KEY"]
+ if os.environ.get("OPENAI_API_BASE") is not None:
+ openai.api_base = os.environ["OPENAI_API_BASE"]
 
  coder = Coder.create(
  main_model,

benchmark/docker.sh

@@ -4,6 +4,7 @@ docker run \
  -it --rm \
  -v `pwd`:/aider \
  -v `pwd`/tmp.benchmarks/.:/benchmarks \
+ -e OPENAI_API_BASE=http://10.0.1.3:5001/v1 \
  -e OPENAI_API_KEY=$OPENAI_API_KEY \
  -e HISTFILE=/aider/.bash_history \
  -e AIDER_DOCKER=1 \

benchmark/rungrid.py

@@ -9,10 +9,11 @@
 def main():
  models = [
  # "gpt-3.5-turbo-0301",
- "gpt-3.5-turbo-0613",
+ #"gpt-3.5-turbo-0613",
  # "gpt-3.5-turbo-16k-0613",
  # "gpt-4-0314",
  # "gpt-4-0613",
+ "starcoderplus",
  ]
  edit_formats = [
  # "diff",
@@ -21,7 +22,7 @@ def main():
  # "whole-func",
  ]
 
- for repeat in range(1, 10, 1):
+ for repeat in range(1, 2, 1):
  for model in models:
  for edit_format in edit_formats:
  # dump(model, edit_format)
@@ -47,7 +48,7 @@ def run(dirname, model, edit_format):
  "--edit-format",
  edit_format,
  "--threads",
- "10",
+ "1",
  "--cont",
  ]
  print(" ".join(cmd))