fix: fixed summary models for deepseek and openrouter, added TODOs for improving openrouter support

gptme/llm/__init__.py

@@ -172,20 +172,21 @@ def _summarize_str(content: str) -> str:
         Message("user", content=f"Summarize this:\n{content}"),
     ]
 
+    # get default provider
     provider: Provider = get_model().provider  # type: ignore
-    model = f"{provider}/{get_summary_model(provider)}"
-    base_model = _get_base_model(model)
-    context_limit = MODELS[provider][base_model]["context"]
 
-    if len_tokens(messages, base_model) > context_limit:
+    # get summary model for provider
+    model = get_model(f"{provider}/{get_summary_model(provider)}")
+
+    if len_tokens(messages, model.model) > model.context:
         raise ValueError(
-            f"Cannot summarize more than {context_limit} tokens, got {len_tokens(messages, base_model)}"
+            f"Cannot summarize more than {model.context} tokens, got {len_tokens(messages, model.model)}"
         )
 
-    summary = _chat_complete(messages, model, None)
+    summary = _chat_complete(messages, model.full, None)
     assert summary
     logger.debug(
-        f"Summarized long output ({len_tokens(content, base_model)} -> {len_tokens(summary, base_model)} tokens): "
+        f"Summarized long output ({len_tokens(content, model.model)} -> {len_tokens(summary, model.model)} tokens): "
         + summary
     )
     return summary

gptme/llm/llm_openai.py

@@ -16,6 +16,7 @@
     from openai import OpenAI  # fmt: skip
     from openai.types.chat import ChatCompletionToolParam  # fmt: skip
 
+
 # Dictionary to store clients for each provider
 clients: dict[Provider, "OpenAI"] = {}
 logger = logging.getLogger(__name__)
@@ -27,6 +28,10 @@
     "X-Title": "gptme",
 }
 
+# TODO: improve provider routing for openrouter: https://openrouter.ai/docs/provider-routing
+# TODO: set required-parameters: https://openrouter.ai/docs/provider-routing#required-parameters-_beta_
+# TODO: set quantization: https://openrouter.ai/docs/provider-routing#quantization
+
 
 ALLOWED_FILE_EXTS = ["jpg", "jpeg", "png", "gif"]
 

gptme/llm/models.py

@@ -258,5 +258,7 @@ def get_summary_model(provider: Provider) -> str:  # pragma: no cover
         return "gemini-1.5-flash-latest"
     elif provider == "anthropic":
         return "claude-3-haiku-20240307"
+    elif provider == "deepseek":
+        return "deepseek-chat"
     else:
         raise ValueError(f"Provider {provider} did not have a summary model")