refactor: Split summarize method and add model metadata handling

aider/history.py

@@ -26,6 +26,12 @@ class ChatSummary:
         return sized
 
     def summarize(self, messages, depth=0):
+        messages = self.summarize_real(messages)
+        if messages and messages[-1]["role"] != "assistant":
+            messages.append(dict(role="assistant", content="Ok."))
+        return messages
+
+    def summarize_real(self, messages, depth=0):
         if not self.models:
             raise ValueError("No models available for summarization")
 
@@ -88,7 +94,7 @@ class ChatSummary:
         if summary_tokens + tail_tokens < self.max_tokens:
             return result
 
-        return self.summarize(result, depth + 1)
+        return self.summarize_real(result, depth + 1)
 
     def summarize_all(self, messages):
         content = ""

aider/models.py

@@ -625,6 +625,7 @@ MODEL_SETTINGS = [
     ModelSettings(
         "deepseek/deepseek-reasoner",
         "diff",
+        weak_model_name="deepseek/deepseek-chat",
         editor_model_name="deepseek/deepseek-chat",
         editor_edit_format="editor-diff",
         use_repo_map=True,

aider/resources/model-metadata.json

@@ -1,2 +1,18 @@
 {
+    "deepseek-reasoner": {
+        "max_tokens": 8192,
+        "max_input_tokens": 128000,
+        "max_output_tokens": 8192,
+        "input_cost_per_token": 0.00000055,
+        "input_cost_per_token_cache_hit": 0.00000014,
+        "cache_read_input_token_cost": 0.00000014,
+        "cache_creation_input_token_cost": 0.0,
+        "output_cost_per_token": 0.00000219,
+        "litellm_provider": "deepseek",
+        "mode": "chat",
+        //"supports_function_calling": true, 
+        "supports_assistant_prefill": true,
+        //"supports_tool_choice": true,
+        "supports_prompt_caching": true
+    },
 }

aider/sendchat.py

@@ -5,6 +5,7 @@ import time
 from aider.dump import dump  # noqa: F401
 from aider.exceptions import LiteLLMExceptions
 from aider.llm import litellm
+from aider.utils import format_messages
 
 # from diskcache import Cache
 
@@ -30,7 +31,9 @@ def sanity_check_messages(messages):
             continue
 
         if last_role and role == last_role:
-            return False
+            print(format_messages(messages))
+            # import sys ; sys.exit()
+            raise ValueError("Messages don't properly alternate user/assistant")
 
         last_role = role
         last_non_system_role = role
@@ -47,6 +50,8 @@ def send_completion(
     temperature=0,
     extra_params=None,
 ):
+    # sanity_check_messages(messages)
+
     kwargs = dict(
         model=model_name,
         messages=messages,

aider/utils.py

@@ -112,7 +112,7 @@ def format_messages(messages, title=None):
         output.append(f"{title.upper()} {'*' * 50}")
 
     for msg in messages:
-        output.append("")
+        output.append("-------")
         role = msg["role"].upper()
         content = msg.get("content")
         if isinstance(content, list):  # Handle list content (e.g., image messages)

benchmark/benchmark.py

@@ -16,6 +16,7 @@ from types import SimpleNamespace
 from typing import List, Optional
 
 import git
+import importlib_resources
 import lox
 import pandas as pd
 import prompts
@@ -737,6 +738,10 @@ def run_test_real(
         chat_history_file=history_fname,
     )
 
+    resource_metadata = importlib_resources.files("aider.resources").joinpath("model-metadata.json")
+    model_metadata_files_loaded = models.register_litellm_models([resource_metadata])
+    dump(model_metadata_files_loaded)
+
     # weak_model_name = model_name
     weak_model_name = None
 
@@ -747,6 +752,8 @@ def run_test_real(
         editor_edit_format=editor_edit_format,
     )
 
+    dump(main_model.max_chat_history_tokens)
+
     if num_ctx:
         if not main_model.extra_params:
             main_model.extra_params = {}