initial code for working with openrouter

aider/models/__init__.py

@@ -0,0 +1,7 @@
+from .openai import OpenAIModel
+from .openrouter import OpenRouterModel
+from .model import Model
+
+GPT4 = Model("gpt-4")
+GPT35 = Model("gpt-3.5-turbo")
+GPT35_16k = Model("gpt-3.5-turbo-16k")

aider/models/model.py

@@ -0,0 +1,60 @@
+import importlib
+
+using_openrouter = False
+
+class Model:
+    name = None
+    edit_format = None
+    max_context_tokens = 0
+    tokenizer = None
+
+    always_available = False
+    use_repo_map = False
+    send_undo_reply = False
+
+    prompt_price = None
+    completion_price = None
+
+    def __init__(self, name, openai=None):
+        global using_openrouter
+        if (openai and "openrouter.ai" in openai.api_base):
+            using_openrouter = True
+
+        from .openai import OpenAIModel
+        from .openrouter import OpenRouterModel
+        model = None
+        if using_openrouter:
+            if name == 'gpt-4':
+                name = 'openai/gpt-4'
+            elif name == 'gpt-3.5-turbo':
+                name = 'openai/gpt-3.5-turbo'
+            elif name == 'gpt-3.5.turbo-16k':
+                name = 'openai/gpt-3.5-turbo-16k'
+
+            model = OpenRouterModel(name, openai)
+        else:
+            model = OpenAIModel(name)
+
+        self.name = model.name
+        self.edit_format = model.edit_format
+        self.max_context_tokens = model.max_context_tokens
+        self.tokenizer = model.tokenizer
+        self.prompt_price = model.prompt_price
+        self.completion_price = model.completion_price
+        self.always_available = model.always_available
+        self.use_repo_map = model.use_repo_map
+
+    def __str__(self):
+        return self.name
+
+    @staticmethod
+    def strong_model():
+        return Model('gpt-4')
+
+    @staticmethod
+    def weak_model():
+        return Model('gpt-3.5-turbo')
+
+    @staticmethod
+    def commit_message_models():
+        return [Model('gpt-3.5-turbo'), Model('gpt-3.5-turbo-16k')]

aider/models/openai.py

@@ -0,0 +1,64 @@
+import tiktoken
+import re
+from .model import Model
+
+known_tokens = {
+    "gpt-3.5-turbo": 4,
+    "gpt-4": 8,
+}
+
+
+class OpenAIModel(Model):
+    def __init__(self, name):
+        self.name = name
+
+        tokens = None
+
+        match = re.search(r"-([0-9]+)k", name)
+        if match:
+            tokens = int(match.group(1))
+        else:
+            for m, t in known_tokens.items():
+                if name.startswith(m):
+                    tokens = t
+
+        if tokens is None:
+            raise ValueError(f"Unknown context window size for model: {name}")
+
+        self.max_context_tokens = tokens * 1024
+        self.tokenizer = tiktoken.encoding_for_model(name)
+
+        if self.is_gpt4():
+            self.edit_format = "diff"
+            self.use_repo_map = True
+            self.send_undo_reply = True
+
+            if tokens == 8:
+                self.prompt_price = 0.03
+                self.completion_price = 0.06
+            elif tokens == 32:
+                self.prompt_price = 0.06
+                self.completion_price = 0.12
+
+            return
+
+        if self.is_gpt35():
+            self.edit_format = "whole"
+            self.always_available = True
+
+            if tokens == 4:
+                self.prompt_price = 0.0015
+                self.completion_price = 0.002
+            elif tokens == 16:
+                self.prompt_price = 0.003
+                self.completion_price = 0.004
+
+            return
+
+        raise ValueError(f"Unsupported model: {name}")
+
+    def is_gpt4(self):
+        return self.name.startswith("gpt-4")
+
+    def is_gpt35(self):
+        return self.name.startswith("gpt-3.5-turbo")

aider/models/openrouter.py

@@ -0,0 +1,13 @@
+import tiktoken
+from .model import Model
+
+
+class OpenRouterModel(Model):
+    def __init__(self, name, openai):
+        self.name = name
+        self.edit_format = "diff"
+        self.use_repo_map = True
+        self.max_context_tokens = 1024 * 8
+
+        # TODO: figure out proper encodings for non openai models
+        self.tokenizer = tiktoken.get_encoding("cl100k_base")