langroid 0.48.3__py3-none-any.whl → 0.49.1__py3-none-any.whl

langroid/language_models/openai_gpt.py

@@ -227,6 +227,16 @@ class OpenAICallParams(BaseModel):
         return {k: v for k, v in self.dict().items() if v is not None}
 
 
+class LiteLLMProxyConfig(BaseSettings):
+    """Configuration for LiteLLM proxy connection."""
+
+    api_key: str = ""  # read from env var LITELLM_API_KEY if set
+    api_base: str = ""  # read from env var LITELLM_API_BASE if set
+
+    class Config:
+        env_prefix = "LITELLM_"
+
+
 class OpenAIGPTConfig(LLMConfig):
     """
     Class for any LLM with an OpenAI-like API: besides the OpenAI models this includes:
@@ -250,6 +260,7 @@ class OpenAIGPTConfig(LLMConfig):
     organization: str = ""
     api_base: str | None = None  # used for local or other non-OpenAI models
     litellm: bool = False  # use litellm api?
+    litellm_proxy: LiteLLMProxyConfig = LiteLLMProxyConfig()
     ollama: bool = False  # use ollama's OpenAI-compatible endpoint?
     min_output_tokens: int = 1
     use_chat_for_completion = True  # do not change this, for OpenAI models!
@@ -520,6 +531,7 @@ class OpenAIGPT(LanguageModel):
         self.is_glhf = self.config.chat_model.startswith("glhf/")
         self.is_openrouter = self.config.chat_model.startswith("openrouter/")
         self.is_langdb = self.config.chat_model.startswith("langdb/")
+        self.is_litellm_proxy = self.config.chat_model.startswith("litellm-proxy/")
 
         if self.is_groq:
             # use groq-specific client
@@ -546,7 +558,14 @@ class OpenAIGPT(LanguageModel):
             )
         else:
             # in these cases, there's no specific client: OpenAI python client suffices
-            if self.is_gemini:
+            if self.is_litellm_proxy:
+                self.config.chat_model = self.config.chat_model.replace(
+                    "litellm-proxy/", ""
+                )
+                if self.api_key == OPENAI_API_KEY:
+                    self.api_key = self.config.litellm_proxy.api_key or self.api_key
+                self.api_base = self.config.litellm_proxy.api_base or self.api_base
+            elif self.is_gemini:
                 self.config.chat_model = self.config.chat_model.replace("gemini/", "")
                 if self.api_key == OPENAI_API_KEY:
                     self.api_key = os.getenv("GEMINI_API_KEY", DUMMY_API_KEY)
@@ -1040,8 +1059,8 @@ class OpenAIGPT(LanguageModel):
                 )
                 if is_break:
                     break
-        except Exception:
-            pass
+        except Exception as e:
+            logging.warning("Error while processing stream response: %s", str(e))
 
         print("")
         # TODO- get usage info in stream mode (?)
@@ -1102,8 +1121,8 @@ class OpenAIGPT(LanguageModel):
                 )
                 if is_break:
                     break
-        except Exception:
-            pass
+        except Exception as e:
+            logging.warning("Error while processing stream response: %s", str(e))
 
         print("")
         # TODO- get usage info in stream mode (?)
@@ -1684,11 +1703,32 @@ class OpenAIGPT(LanguageModel):
             if self.config.litellm and settings.debug:
                 kwargs["logger_fn"] = litellm_logging_fn
             result = completion_call(**kwargs)
-            if not self.get_stream():
-                # if streaming, cannot cache result
+
+            if self.get_stream():
+                # If streaming, cannot cache result
                 # since it is a generator. Instead,
                 # we hold on to the hashed_key and
                 # cache the result later
+
+                # Test if this is a stream with an exception by
+                # trying to get first chunk: Some providers like LiteLLM
+                # produce a valid stream object `result` instead of throwing a
+                # rate-limit error, and if we don't catch it here,
+                # we end up returning an empty response and not
+                # using the retry mechanism in the decorator.
+                try:
+                    # try to get the first chunk to check for errors
+                    test_iter = iter(result)
+                    first_chunk = next(test_iter)
+                    # If we get here without error, recreate the stream
+                    result = chain([first_chunk], test_iter)
+                except StopIteration:
+                    # Empty stream is fine
+                    pass
+                except Exception as e:
+                    # Propagate any errors in the stream
+                    raise e
+            else:
                 self._cache_store(hashed_key, result.model_dump())
         return cached, hashed_key, result
 
@@ -1715,7 +1755,35 @@ class OpenAIGPT(LanguageModel):
                 kwargs["logger_fn"] = litellm_logging_fn
             # If it's not in the cache, call the API
             result = await acompletion_call(**kwargs)
-            if not self.get_stream():
+            if self.get_stream():
+                try:
+                    # Try to peek at the first chunk to immediately catch any errors
+                    # Store the original result (the stream)
+                    original_stream = result
+
+                    # Manually create and advance the iterator to check for errors
+                    stream_iter = original_stream.__aiter__()
+                    try:
+                        # This will raise an exception if the stream is invalid
+                        first_chunk = await anext(stream_iter)
+
+                        # If we reach here, the stream started successfully
+                        # Now recreate a fresh stream from the original API result
+                        # Otherwise, return a new stream that yields the first chunk
+                        # and remaining items
+                        async def combined_stream():  # type: ignore
+                            yield first_chunk
+                            async for chunk in stream_iter:
+                                yield chunk
+
+                        result = combined_stream()  # type: ignore
+                    except StopAsyncIteration:
+                        # Empty stream is normal - nothing to do
+                        pass
+                except Exception as e:
+                    # Any exception here should be raised to trigger the retry mechanism
+                    raise e
+            else:
                 self._cache_store(hashed_key, result.model_dump())
         return cached, hashed_key, result
 

langroid/mytypes.py

@@ -65,6 +65,30 @@ class DocMetaData(BaseModel):
 
         return original_dict
 
+    def __str__(self) -> str:
+        title_str = (
+            ""
+            if "unknown" in self.title.lower() or self.title.strip() == ""
+            else f"Title: {self.title}"
+        )
+        date_str = ""
+        if (
+            "unknown" not in self.published_date.lower()
+            and self.published_date.strip() != ""
+        ):
+            try:
+                from dateutil import parser
+
+                # Try to parse the date string
+                date_obj = parser.parse(self.published_date)
+                # Format to include only the date part (year-month-day)
+                date_only = date_obj.strftime("%Y-%m-%d")
+                date_str = f"Date: {date_only}"
+            except (ValueError, ImportError, TypeError):
+                # If parsing fails, just use the original date
+                date_str = f"Date: {self.published_date}"
+        return f"{self.source} {title_str} {date_str}".strip()
+
     class Config:
         extra = Extra.allow
 
@@ -93,7 +117,7 @@ class Document(BaseModel):
         return dedent(
             f"""
         CONTENT: {self.content}         
-        SOURCE:{self.metadata.source}
+        SOURCE:{str(self.metadata)}
         """
         )
 

langroid/utils/output/citations.py

@@ -97,7 +97,7 @@ def format_cited_references(
         # source and content for each citation
         full_citations_str = "\n".join(
             [
-                f"[^{c}] {passages[c-1].metadata.source}"
+                f"[^{c}] {str(passages[c-1].metadata)}"
                 f"\n{format_footnote_text(passages[c-1].content)}"
                 for c in good_citations
             ]
@@ -105,6 +105,6 @@ def format_cited_references(
 
         # source for each citation
         citations_str = "\n".join(
-            [f"[^{c}] {passages[c-1].metadata.source}" for c in good_citations]
+            [f"[^{c}] {str(passages[c-1].metadata)}" for c in good_citations]
         )
     return full_citations_str, citations_str

{langroid-0.48.3.dist-info → langroid-0.49.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: langroid
-Version: 0.48.3
+Version: 0.49.1
 Summary: Harness LLMs with Multi-Agent Programming
 Author-email: Prasad Chalasani <pchalasani@gmail.com>
 License: MIT

{langroid-0.48.3.dist-info → langroid-0.49.1.dist-info}/RECORD

@@ -1,6 +1,6 @@
 langroid/__init__.py,sha256=z_fCOLQJPOw3LLRPBlFB5-2HyCjpPgQa4m4iY5Fvb8Y,1800
 langroid/exceptions.py,sha256=OPjece_8cwg94DLPcOGA1ddzy5bGh65pxzcHMnssTz8,2995
-langroid/mytypes.py,sha256=1mt21xiczAu1WgfZIptBKN9WpqYgpVeo8cAXvs6drKA,2986
+langroid/mytypes.py,sha256=ezj_6FFDkJZiVx1SS9eJvh23dH76Ti7mJbePi8ldkAI,3919
 langroid/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 langroid/agent/__init__.py,sha256=ll0Cubd2DZ-fsCMl7e10hf9ZjFGKzphfBco396IKITY,786
 langroid/agent/base.py,sha256=U-UjdpxIFqkzRIB5-LYwHrhMSNI3sDbfnNRqIhrtsyI,79568
@@ -72,7 +72,7 @@ langroid/language_models/base.py,sha256=mDYmFCBCLdq8_Uvws4MiewwEgcOCP8Qb0e5yUXr3
 langroid/language_models/config.py,sha256=9Q8wk5a7RQr8LGMT_0WkpjY8S4ywK06SalVRjXlfCiI,378
 langroid/language_models/mock_lm.py,sha256=5BgHKDVRWFbUwDT_PFgTZXz9-k8wJSA2e3PZmyDgQ1k,4022
 langroid/language_models/model_info.py,sha256=tfBBxL0iUf2mVN6CjcvqflzFUVg2oZqOJZexZ8jHTYA,12216
-langroid/language_models/openai_gpt.py,sha256=JJXLw7iuQFbW0CkcKiYjX4agRllRhDaGT9Q8OCdEK-8,79559
+langroid/language_models/openai_gpt.py,sha256=M_jp97Ksp5r3U-d0jCLPLjVmn7IK1mC8Ry4t7k6A5tc,82906
 langroid/language_models/utils.py,sha256=L4_CbihDMTGcsg0TOG1Yd5JFEto46--h7CX_14m89sQ,5016
 langroid/language_models/prompt_formatter/__init__.py,sha256=2-5cdE24XoFDhifOLl8yiscohil1ogbP1ECkYdBlBsk,372
 langroid/language_models/prompt_formatter/base.py,sha256=eDS1sgRNZVnoajwV_ZIha6cba5Dt8xjgzdRbPITwx3Q,1221
@@ -115,7 +115,7 @@ langroid/utils/types.py,sha256=-BvyIf_LmAJ5jR9NC7S4CSVNEr3XayAaxJ5o0TiIej0,2992
 langroid/utils/algorithms/__init__.py,sha256=WylYoZymA0fnzpB4vrsH_0n7WsoLhmuZq8qxsOCjUpM,41
 langroid/utils/algorithms/graph.py,sha256=JbdpPnUOhw4-D6O7ou101JLA3xPCD0Lr3qaPoFCaRfo,2866
 langroid/utils/output/__init__.py,sha256=7P0f--4IZneNsTxXY5fd6d6iW-CeVe-KSsl-87sbBPc,340
-langroid/utils/output/citations.py,sha256=5mhg2-kpBF7qgV82gJbIiUHfoJcpJiK1sAWdDF0o0Wc,3866
+langroid/utils/output/citations.py,sha256=cEiqSH7DJ5q4M2z_6eFjCj9Ohnf68i6sivjeRFuFAtk,3862
 langroid/utils/output/printing.py,sha256=yzPJZN-8_jyOJmI9N_oLwEDfjMwVgk3IDiwnZ4eK_AE,2962
 langroid/utils/output/status.py,sha256=rzbE7mDJcgNNvdtylCseQcPGCGghtJvVq3lB-OPJ49E,1049
 langroid/vector_store/__init__.py,sha256=8ktJUVsVUoc7FMmkUFpFBZu7VMWUqQY9zpm4kEJ8yTs,1537
@@ -127,7 +127,7 @@ langroid/vector_store/pineconedb.py,sha256=otxXZNaBKb9f_H75HTaU3lMHiaR2NUp5MqwLZ
 langroid/vector_store/postgres.py,sha256=wHPtIi2qM4fhO4pMQr95pz1ZCe7dTb2hxl4VYspGZoA,16104
 langroid/vector_store/qdrantdb.py,sha256=O6dSBoDZ0jzfeVBd7LLvsXu083xs2fxXtPa9gGX3JX4,18443
 langroid/vector_store/weaviatedb.py,sha256=Yn8pg139gOy3zkaPfoTbMXEEBCiLiYa1MU5d_3UA1K4,11847
-langroid-0.48.3.dist-info/METADATA,sha256=Dhbb_Of_lBJcDtv1ezx1zjTkuW5jFi2NegQ4LA_fRkU,63606
-langroid-0.48.3.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-langroid-0.48.3.dist-info/licenses/LICENSE,sha256=EgVbvA6VSYgUlvC3RvPKehSg7MFaxWDsFuzLOsPPfJg,1065
-langroid-0.48.3.dist-info/RECORD,,
+langroid-0.49.1.dist-info/METADATA,sha256=a2cArSN5YfRq4GRH37MkO6h-fvXbXEFkoo-qDMyVTzA,63606
+langroid-0.49.1.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+langroid-0.49.1.dist-info/licenses/LICENSE,sha256=EgVbvA6VSYgUlvC3RvPKehSg7MFaxWDsFuzLOsPPfJg,1065
+langroid-0.49.1.dist-info/RECORD,,