fix: fixed support for groq (and deepseek?) (#231)

gptme/llm_anthropic.py

@@ -38,7 +38,7 @@ class MessagePart(TypedDict, total=False):
 def chat(messages: list[Message], model: str) -> str:
     assert anthropic, "LLM not initialized"
     messages, system_messages = _transform_system_messages(messages)
-    messages_dicts = msgs2dicts(messages, anthropic=True)
+    messages_dicts = msgs2dicts(messages, provider="anthropic")
     response = anthropic.beta.prompt_caching.messages.create(
         model=model,
         messages=messages_dicts,  # type: ignore
@@ -56,7 +56,7 @@ def chat(messages: list[Message], model: str) -> str:
 def stream(messages: list[Message], model: str) -> Generator[str, None, None]:
     assert anthropic, "LLM not initialized"
     messages, system_messages = _transform_system_messages(messages)
-    messages_dicts = msgs2dicts(messages, anthropic=True)
+    messages_dicts = msgs2dicts(messages, provider="anthropic")
     with anthropic.beta.prompt_caching.messages.stream(
         model=model,
         messages=messages_dicts,  # type: ignore

gptme/llm_openai.py

@@ -60,6 +60,20 @@ def init(provider: Provider, config: Config):
     assert openai, "Provider not initialized"
 
 
+def get_provider() -> Provider | None:
+    if not openai:
+        return None
+    if "openai.com" in str(openai.base_url):
+        return "openai"
+    if "openrouter.ai" in str(openai.base_url):
+        return "openrouter"
+    if "groq.com" in str(openai.base_url):
+        return "groq"
+    if "x.ai" in str(openai.base_url):
+        return "xai"
+    return None
+
+
 def get_client() -> "OpenAI | None":
     return openai
 
@@ -88,7 +102,7 @@ def chat(messages: list[Message], model: str) -> str:
 
     response = openai.chat.completions.create(
         model=model,
-        messages=msgs2dicts(messages, openai=True),  # type: ignore
+        messages=msgs2dicts(messages, provider=get_provider()),  # type: ignore
         temperature=TEMPERATURE if not is_o1 else NOT_GIVEN,
         top_p=TOP_P if not is_o1 else NOT_GIVEN,
         extra_headers=(
@@ -105,7 +119,7 @@ def stream(messages: list[Message], model: str) -> Generator[str, None, None]:
     stop_reason = None
     for chunk in openai.chat.completions.create(
         model=model,
-        messages=msgs2dicts(_prep_o1(messages), openai=True),  # type: ignore
+        messages=msgs2dicts(_prep_o1(messages), provider=get_provider()),  # type: ignore
         temperature=TEMPERATURE,
         top_p=TOP_P,
         stream=True,

gptme/message.py

@@ -16,6 +16,7 @@
 
 from .codeblock import Codeblock
 from .constants import ROLE_COLOR
+from .models import Provider
 from .util import console, get_tokenizer, rich_to_str
 
 logger = logging.getLogger(__name__)
@@ -26,6 +27,9 @@
 max_system_len = 20000
 
 
+ProvidersWithFiles: list[Provider] = ["openai", "anthropic", "openrouter"]
+
+
 @dataclass(frozen=True, eq=False)
 class Message:
     """
@@ -75,10 +79,11 @@ def replace(self, **kwargs) -> Self:
         return dataclasses.replace(self, **kwargs)
 
     def _content_files_list(
-        self, openai: bool = False, anthropic: bool = False
+        self,
+        provider: Provider,
     ) -> list[dict[str, Any]]:
         # only these providers support files in the content
-        if not openai and not anthropic:
+        if provider not in ProvidersWithFiles:
             raise ValueError("Provider does not support files in the content")
 
         # combines a content message with a list of files
@@ -121,7 +126,7 @@ def _content_files_list(
                 )
                 continue
 
-            if anthropic:
+            if provider == "anthropic":
                 content.append(
                     {
                         "type": "image",
@@ -132,7 +137,7 @@ def _content_files_list(
                         },
                     }
                 )
-            elif openai:
+            elif provider == "openai":
                 # OpenAI format
                 content.append(
                     {
@@ -147,12 +152,13 @@ def _content_files_list(
 
         return content
 
-    def to_dict(self, keys=None, openai=False, anthropic=False) -> dict:
+    def to_dict(self, keys=None, provider: Provider | None = None) -> dict:
         """Return a dict representation of the message, serializable to JSON."""
         content: str | list[dict[str, Any]]
-        if anthropic or openai:
-            # OpenAI format or Anthropic format should include files in the content
-            content = self._content_files_list(openai=openai, anthropic=anthropic)
+        if provider in ProvidersWithFiles:
+            # OpenAI/Anthropic format should include files in the content
+            # Some otherwise OpenAI-compatible providers (groq, deepseek?) do not support this
+            content = self._content_files_list(provider)
         else:
             # storage/wire format should keep the content as a string
             content = self.content
@@ -346,12 +352,9 @@ def toml_to_msgs(toml: str) -> list[Message]:
     ]
 
 
-def msgs2dicts(msgs: list[Message], openai=False, anthropic=False) -> list[dict]:
+def msgs2dicts(msgs: list[Message], provider: Provider) -> list[dict]:
     """Convert a list of Message objects to a list of dicts ready to pass to an LLM."""
-    return [
-        msg.to_dict(keys=["role", "content"], openai=openai, anthropic=anthropic)
-        for msg in msgs
-    ]
+    return [msg.to_dict(keys=["role", "content"], provider=provider) for msg in msgs]
 
 
 # TODO: remove model assumption