More flexible models

aider/coder.py

@@ -79,19 +79,20 @@ class Coder:
             self.console = Console(force_terminal=True, no_color=True)
 
         main_model = models.get_model(main_model)
-        if main_model not in models.GPT35_models:
+        if not main_model.is_always_available():
             if not self.check_model_availability(main_model):
                 if main_model != models.GPT4:
                     self.io.tool_error(f"API key does not support {main_model.name}.")
                 main_model = models.GPT35_16k
 
         self.main_model = main_model
-        if main_model in models.GPT35_models:
+        if main_model.is_gpt35():
             self.io.tool_output(
                 f"Using {main_model.name} (experimental): disabling ctags/repo-maps.",
             )
             self.gpt_prompts = prompts.GPT35()
         else:
+            self.io.tool_output(f"Using {main_model.name}.")
             self.gpt_prompts = prompts.GPT4()
 
         self.show_diffs = show_diffs
@@ -107,7 +108,7 @@ class Coder:
             self.io.tool_output("Not using git.")
             self.find_common_root()
 
-        if main_model in models.GPT4_models:
+        if main_model.is_gpt4():
             rm_io = io if self.verbose else None
             self.repo_map = RepoMap(
                 map_tokens,
@@ -318,7 +319,7 @@ class Coder:
         ]
 
         main_sys = self.gpt_prompts.main_system
-        if self.main_model in models.GPT4_models:
+        if self.main_model.is_gpt4():
             main_sys += "\n" + self.gpt_prompts.system_reminder
 
         messages = [
@@ -347,7 +348,7 @@ class Coder:
         if edit_error:
             return edit_error
 
-        if self.main_model in models.GPT4_models or not edited:
+        if self.main_model.is_gpt4() or not edited:
             # Don't add 3.5 assistant messages to the history if they contain "edits"
             # Because those edits are actually fully copies of the file!
             # That wastes too much context window.
@@ -482,7 +483,7 @@ class Coder:
 
                 if self.pretty:
                     show_resp = self.resp
-                    if self.main_model in models.GPT35_models:
+                    if self.main_model.is_gpt35():
                         try:
                             show_resp = self.update_files_gpt35(self.resp, mode="diff")
                         except ValueError:
@@ -786,9 +787,9 @@ class Coder:
         return set(self.get_all_relative_files()) - set(self.get_inchat_relative_files())
 
     def apply_updates(self, content):
-        if self.main_model in models.GPT4_models:
+        if self.main_model.is_gpt4():
             method = self.update_files_gpt4
-        elif self.main_model in models.GPT35_models:
+        elif self.main_model.is_gpt35():
             method = self.update_files_gpt35
         else:
             raise ValueError(f"apply_updates() doesn't support {self.main_model.name}")

aider/commands.py

@@ -8,7 +8,7 @@ import git
 import tiktoken
 from prompt_toolkit.completion import Completion
 
-from aider import models, prompts, utils
+from aider import prompts, utils
 
 
 class Commands:
@@ -183,7 +183,7 @@ class Commands:
             "was reset and removed from git.\n"
         )
 
-        if self.coder.main_model in models.GPT4_models:
+        if self.coder.main_model.is_gpt4():
             return prompts.undo_command_reply
 
     def cmd_diff(self, args):

aider/main.py

@@ -84,7 +84,7 @@ def main(args=None, input=None, output=None):
         action="store_const",
         dest="model",
         const=models.GPT35_16k.name,
-        help=f"Use {models.GPT35.name} model for the main chat (not advised)",
+        help=f"Use {models.GPT35_16k.name} model for the main chat (gpt-4 is better)",
     )
     parser.add_argument(
         "--pretty",

aider/models.py

@@ -1,44 +1,37 @@
-class Model:
-    def __init__(self, name, max_context_tokens):
-        self.name = name
-        self.max_context_tokens = max_context_tokens * 1024
-
-
-# 4
-
-GPT4_32k = Model("gpt-4-32k", 32)
-GPT4_32k_0613 = Model("gpt-4-32k-0613", 32)
-GPT4 = Model("gpt-4", 8)
-
-GPT4_models = [GPT4, GPT4_32k, GPT4_32k_0613]
-
-# 3.5
-
-GPT35 = Model("gpt-3.5-turbo", 4)
-GPT35_16k = Model("gpt-3.5-turbo-16k", 16)
-
-GPT35_models = [GPT35, GPT35_16k]
-
-
 import re
 
+
+class Model:
+    def __init__(self, name, tokens=None):
+        self.name = name
+        if tokens is None:
+            match = re.search(r"-([0-9]+)k", name)
+
+            default_tokens = 8
+
+            tokens = int(match.group(1)) if match else default_tokens
+
+        self.max_context_tokens = tokens * 1024
+
+    def is_gpt4(self):
+        return self.name.startswith("gpt-4")
+
+    def is_gpt35(self):
+        return self.name.startswith("gpt-3.5-turbo")
+
+    def is_always_available(self):
+        return self.is_gpt35()
+
+
+GPT4 = Model("gpt-4", 8)
+GPT35 = Model("gpt-3.5-turbo")
+GPT35_16k = Model("gpt-3.5-turbo-16k")
+
+
 def get_model(name):
-    models = GPT35_models + GPT4_models
+    model = Model(name)
 
-    for model in models:
-        if model.name == name:
+    if model.is_gpt4() or model.is_gpt35():
         return model
 
-    match = re.search(r'-([0-9]+)k', name)
-    tokens = int(match.group(1)) if match else 0
-
-    model = Model(name, tokens)
-
-    if name.startswith("gpt-4-"):
-        GPT4_models.append(model)
-    elif name.startswith("gpt-3.5-"):
-        GPT35_models.append(model)
-    else:
     raise ValueError(f"Unsupported model: {name}")
-
-    return model