langroid 0.52.6__py3-none-any.whl → 0.52.8__py3-none-any.whl

langroid/agent/special/doc_chat_agent.py

@@ -872,6 +872,7 @@ class DocChatAgent(ChatAgent):
         else:
             answer_doc = super().llm_response_forget(final_prompt)
 
+        assert answer_doc is not None, "LLM response should not be None here"
         final_answer = answer_doc.content.strip()
         show_if_debug(final_answer, "SUMMARIZE_RESPONSE= ")
 

langroid/language_models/openai_gpt.py

@@ -336,15 +336,16 @@ class OpenAIGPTConfig(LLMConfig):
         # (e.g. anthropic doesn't like first msg to be system msg)
         litellm.modify_params = True
         self.seed = None  # some local mdls don't support seed
-        keys_dict = litellm.utils.validate_environment(self.chat_model)
-        missing_keys = keys_dict.get("missing_keys", [])
-        if len(missing_keys) > 0:
-            raise ValueError(
-                f"""
-                Missing environment variables for litellm-proxied model:
-                {missing_keys}
-                """
-            )
+        if self.api_key == DUMMY_API_KEY:
+            keys_dict = litellm.utils.validate_environment(self.chat_model)
+            missing_keys = keys_dict.get("missing_keys", [])
+            if len(missing_keys) > 0:
+                raise ValueError(
+                    f"""
+                    Missing environment variables for litellm-proxied model:
+                    {missing_keys}
+                    """
+                )
 
     @classmethod
     def create(cls, prefix: str) -> Type["OpenAIGPTConfig"]:
@@ -1504,6 +1505,9 @@ class OpenAIGPT(LanguageModel):
                     from litellm import completion as litellm_completion
 
                     completion_call = litellm_completion
+
+                    if self.api_key != DUMMY_API_KEY:
+                        kwargs["api_key"] = self.api_key
                 else:
                     if self.client is None:
                         raise ValueError(
@@ -1587,6 +1591,10 @@ class OpenAIGPT(LanguageModel):
             else:
                 if self.config.litellm:
                     from litellm import acompletion as litellm_acompletion
+
+                    if self.api_key != DUMMY_API_KEY:
+                        kwargs["api_key"] = self.api_key
+
                 # TODO this may not work: text_completion is not async,
                 # and we didn't find an async version in litellm
                 assert isinstance(self.async_client, AsyncOpenAI)
@@ -1734,6 +1742,9 @@ class OpenAIGPT(LanguageModel):
                 from litellm import completion as litellm_completion
 
                 completion_call = litellm_completion
+
+                if self.api_key != DUMMY_API_KEY:
+                    kwargs["api_key"] = self.api_key
             else:
                 if self.client is None:
                     raise ValueError("OpenAI/equivalent chat-completion client not set")
@@ -1792,6 +1803,9 @@ class OpenAIGPT(LanguageModel):
                 from litellm import acompletion as litellm_acompletion
 
                 acompletion_call = litellm_acompletion
+
+                if self.api_key != DUMMY_API_KEY:
+                    kwargs["api_key"] = self.api_key
             else:
                 if self.async_client is None:
                     raise ValueError(
@@ -1893,7 +1907,8 @@ class OpenAIGPT(LanguageModel):
             max_completion_tokens=max_tokens,
             stream=self.get_stream(),
         )
-        if self.get_stream():
+        if self.get_stream() and "groq" not in self.chat_model_orig:
+            # groq fails when we include stream_options in the request
             args.update(
                 dict(
                     # get token-usage numbers in stream mode from OpenAI API,

langroid/language_models/utils.py

@@ -67,6 +67,7 @@ def retry_with_exponential_backoff(
                     for err in [
                         "BadRequestError",
                         "ConnectionError",
+                        "NotFoundError",
                     ]
                 ):
                     logger.error(f"OpenAI API request failed with error: {e}.")
@@ -137,6 +138,19 @@ def async_retry_with_exponential_backoff(
                 raise e
             # Retry on specified errors
             except errors as e:
+                # For certain types of errors that slip through here
+                # (e.g. when using proxies like LiteLLM, do not retry)
+                if any(
+                    err in str(e)
+                    for err in [
+                        "BadRequestError",
+                        "ConnectionError",
+                        "NotFoundError",
+                    ]
+                ):
+                    logger.error(f"OpenAI API request failed with error: {e}.")
+                    raise e
+
                 # Increment retries
                 num_retries += 1
 

langroid/parsing/pdf_utils.py

@@ -2,7 +2,7 @@ import tempfile
 from io import BytesIO
 from pathlib import Path
 from tempfile import TemporaryDirectory
-from typing import TYPE_CHECKING, Any, BinaryIO, List, Tuple, Union
+from typing import TYPE_CHECKING, Any, BinaryIO, List, Optional, Tuple, Union
 
 try:
     import fitz
@@ -18,16 +18,23 @@ if fitz is None:
 
 def pdf_split_pages(
     input_pdf: Union[BytesIO, BinaryIO, str],
+    splits: Optional[List[int]] = None,
 ) -> Tuple[List[Path], TemporaryDirectory[Any]]:
-    """Splits a PDF into individual pages in a temporary directory.
+    """Splits a PDF into individual pages or chunks in a temporary directory.
 
     Args:
         input_pdf: Input PDF file in bytes, binary mode, or a file path
+        splits: Optional list of page numbers to split at.
+                If provided, pages will be grouped into chunks ending at
+                these page numbers.
+                For example, if splits = [4, 9], the result will have pages 1-4, 5-9,
+                and 10-end.
+                If not provided, default to splitting into individual pages.
         max_workers: Maximum number of concurrent workers for parallel processing
 
     Returns:
         Tuple containing:
-            - List of paths to individual PDF pages
+            - List of paths to individual PDF pages or chunks
             - Temporary directory object (caller must call cleanup())
 
     Example:
@@ -42,13 +49,41 @@ def pdf_split_pages(
         doc = fitz.open(stream=input_pdf, filetype="pdf")
     paths = []
 
-    for page_num in range(len(doc)):
-        new_doc = fitz.open()
-        new_doc.insert_pdf(doc, from_page=page_num, to_page=page_num)
-        output = Path(tmp_dir.name) / f"page_{page_num + 1}.pdf"
-        new_doc.save(str(output))
-        new_doc.close()
-        paths.append(output)
+    total_pages = len(doc)
+
+    if splits is None:
+        # Split into individual pages (original behavior)
+        for page_num in range(total_pages):
+            new_doc = fitz.open()
+            new_doc.insert_pdf(doc, from_page=page_num, to_page=page_num)
+            output = Path(tmp_dir.name) / f"page_{page_num + 1}.pdf"
+            new_doc.save(str(output))
+            new_doc.close()
+            paths.append(output)
+    else:
+        # Split according to specified page ranges
+        # Make sure the splits list is sorted and includes all valid splits
+        splits = sorted([s for s in splits if 1 <= s <= total_pages])
+
+        # Create the ranges to process
+        ranges = []
+        start_page = 0
+        for end_page in splits:
+            ranges.append((start_page, end_page - 1))
+            start_page = end_page
+
+        # Add the final range if there are pages after the last split
+        if start_page < total_pages:
+            ranges.append((start_page, total_pages - 1))
+
+        # Process each range
+        for i, (from_page, to_page) in enumerate(ranges):
+            new_doc = fitz.open()
+            new_doc.insert_pdf(doc, from_page=from_page, to_page=to_page)
+            output = Path(tmp_dir.name) / f"pages_{from_page + 1}_to_{to_page + 1}.pdf"
+            new_doc.save(str(output))
+            new_doc.close()
+            paths.append(output)
 
     doc.close()
     return paths, tmp_dir

{langroid-0.52.6.dist-info → langroid-0.52.8.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: langroid
-Version: 0.52.6
+Version: 0.52.8
 Summary: Harness LLMs with Multi-Agent Programming
 Author-email: Prasad Chalasani <pchalasani@gmail.com>
 License: MIT

{langroid-0.52.6.dist-info → langroid-0.52.8.dist-info}/RECORD

@@ -14,7 +14,7 @@ langroid/agent/xml_tool_message.py,sha256=6SshYZJKIfi4mkE-gIoSwjkEYekQ8GwcSiCv7a
 langroid/agent/callbacks/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 langroid/agent/callbacks/chainlit.py,sha256=UHB6P_J40vsVnssosqkpkOVWRf9NK4TOY0_G2g_Arsg,20900
 langroid/agent/special/__init__.py,sha256=gik_Xtm_zV7U9s30Mn8UX3Gyuy4jTjQe9zjiE3HWmEo,1273
-langroid/agent/special/doc_chat_agent.py,sha256=-ABgZiaIowFVqYMYlU1Nf8WfOqXOTmlMCEnafEXucPo,65564
+langroid/agent/special/doc_chat_agent.py,sha256=7PvVKHrXHw2LoSgU2-3hE7mz46r5oKB3o_bFhWmfT_I,65642
 langroid/agent/special/doc_chat_task.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 langroid/agent/special/lance_doc_chat_agent.py,sha256=s8xoRs0gGaFtDYFUSIRchsgDVbS5Q3C2b2mr3V1Fd-Q,10419
 langroid/agent/special/lance_tools.py,sha256=qS8x4wi8mrqfbYV2ztFzrcxyhHQ0ZWOc-zkYiH7awj0,2105
@@ -72,8 +72,8 @@ langroid/language_models/base.py,sha256=pfN3t-BktKmN_4K8pwmpjC9OdcHxsytM5s5TmsJ-
 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=vOaTi-XFKnz-BvHUvgjnt0XfOtl21Apev3Zy7Rhckbw,14458
-langroid/language_models/openai_gpt.py,sha256=Hw4llY9Dbox7L0sVP06q10ayoem8FOfXSfbLm3u55v4,85317
-langroid/language_models/utils.py,sha256=hC5p61P_Qlrowkm5wMap1A1b5ZUCwK_XhPIzAQk1T1s,5483
+langroid/language_models/openai_gpt.py,sha256=F28jqTEerN32m14q3K0oc3vnvBT8J7Q9xqXGZNKUjKU,85938
+langroid/language_models/utils.py,sha256=n55Oe2_V_4VNGhytvPWLYC-0tFS07RTjN83KWl-p_MI,6032
 langroid/language_models/prompt_formatter/__init__.py,sha256=2-5cdE24XoFDhifOLl8yiscohil1ogbP1ECkYdBlBsk,372
 langroid/language_models/prompt_formatter/base.py,sha256=eDS1sgRNZVnoajwV_ZIha6cba5Dt8xjgzdRbPITwx3Q,1221
 langroid/language_models/prompt_formatter/hf_formatter.py,sha256=PVJppmjRvD-2DF-XNC6mE05vTZ9wbu37SmXwZBQhad0,5055
@@ -87,7 +87,7 @@ langroid/parsing/md_parser.py,sha256=JUgsUpCaeAuBndmtDaJR9HMZaje1gmtXtaLXJHst3i8
 langroid/parsing/para_sentence_split.py,sha256=AJBzZojP3zpB-_IMiiHismhqcvkrVBQ3ZINoQyx_bE4,2000
 langroid/parsing/parse_json.py,sha256=aADo38bAHQhC8on4aWZZzVzSDy-dK35vRLZsFI2ewh8,4756
 langroid/parsing/parser.py,sha256=uaAITarcGI2504zcP_dLhp3LjNdh9A6R_yS-o_VcaH8,15599
-langroid/parsing/pdf_utils.py,sha256=9HnwhbZvpBUhW8WjY9OpGPKaIt3oe_a1AuqhWKqNQ6s,1616
+langroid/parsing/pdf_utils.py,sha256=QogxU_B1N3WSLyZ9PEcJDaJoZShKs7CPQRVyF1V2DiE,3143
 langroid/parsing/repo_loader.py,sha256=NpysuyzRHvgL3F4BB_wGo5sCUnZ3FOlVCJmZ7CaUdbs,30202
 langroid/parsing/routing.py,sha256=-FcnlqldzL4ZoxuDwXjQPNHgBe9F9-F4R6q7b_z9CvI,1232
 langroid/parsing/search.py,sha256=0NJ5-Rou_BbrHAD7O9b20bKjZJnbadjObvGm4Zq8Kis,9818
@@ -129,7 +129,7 @@ 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-0.52.6.dist-info/METADATA,sha256=fOf2NTPNfbS1F9rSfa7oLtYHB66xFiN65MnRWybC9sU,63519
-langroid-0.52.6.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-langroid-0.52.6.dist-info/licenses/LICENSE,sha256=EgVbvA6VSYgUlvC3RvPKehSg7MFaxWDsFuzLOsPPfJg,1065
-langroid-0.52.6.dist-info/RECORD,,
+langroid-0.52.8.dist-info/METADATA,sha256=v0NjlHl9pXUfWPYH4-Gobzo4fYN9fcCpDaFG3cPFX-A,63519
+langroid-0.52.8.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+langroid-0.52.8.dist-info/licenses/LICENSE,sha256=EgVbvA6VSYgUlvC3RvPKehSg7MFaxWDsFuzLOsPPfJg,1065
+langroid-0.52.8.dist-info/RECORD,,