feaat: Updated qdrant client to 1.9.1, bumped sdk to 0.25.0 (#46)

Updated qdrant client to 1.9.1, bumped sdk to 0.25.0

Author: chandrasekharan-zipstack

pdm.lock

@@ -12,10 +12,12 @@ dependencies = [
     "python-magic~=0.4.27",
     "python-dotenv==1.0.0",
     # LLM Triad
-    "unstract-adapters~=0.12.1",
+    "unstract-adapters~=0.14.0",
     "llama-index==0.10.28",
     "tiktoken~=0.4.0",
     "transformers==4.37.0",
+    # Error handling, remove after moving it to adapters
+    "openai~=1.21.2"
 ]
 readme = "README.md"
 urls = { Homepage = "https://unstract.com", "Release notes" = "https://github.com/Zipstack/unstract-sdk/releases", Source = "https://github.com/Zipstack/unstract-sdk" }

pyproject.toml

@@ -1,4 +1,4 @@
-__version__ = "0.24.1"
+__version__ = "0.25.0"
 
 
 def get_sdk_version():

src/unstract/sdk/__init__.py

@@ -4,7 +4,7 @@
 
 from unstract.sdk.adapters import ToolAdapter
 from unstract.sdk.constants import LogLevel
-from unstract.sdk.exceptions import SdkError
+from unstract.sdk.exceptions import SdkError, ToolEmbeddingError
 from unstract.sdk.tool.base import BaseTool
 
 
@@ -32,23 +32,20 @@ def get_embedding(self, adapter_instance_id: str) -> BaseEmbedding:
             embedding_adapter_id = embedding_config_data.get(Common.ADAPTER_ID)
             if embedding_adapter_id not in self.embedding_adapters:
                 raise SdkError(
-                    f"Embedding adapter not supported : "
-                    f"{embedding_adapter_id}"
+                    f"Embedding adapter not supported : " f"{embedding_adapter_id}"
                 )
 
             embedding_adapter = self.embedding_adapters[embedding_adapter_id][
                 Common.METADATA
             ][Common.ADAPTER]
-            embedding_metadata = embedding_config_data.get(
-                Common.ADAPTER_METADATA
-            )
+            embedding_metadata = embedding_config_data.get(Common.ADAPTER_METADATA)
             embedding_adapter_class = embedding_adapter(embedding_metadata)
             return embedding_adapter_class.get_embedding_instance()
         except Exception as e:
             self.tool.stream_log(
                 log=f"Error getting embedding: {e}", level=LogLevel.ERROR
             )
-            raise SdkError(f"Error getting embedding instance: {e}")
+            raise ToolEmbeddingError(f"Error getting embedding instance: {e}")
 
     def get_embedding_length(self, embedding: BaseEmbedding) -> int:
         embedding_list = embedding._get_text_embedding(self.__TEST_SNIPPET)

src/unstract/sdk/embedding.py

@@ -15,3 +15,19 @@ def __init__(self, message: str = ""):
         if "404" in message:
             message = "Index not found. Please check vector db adapter settings."
         super().__init__(message)
+
+
+class ToolLLMError(SdkError):
+    DEFAULT_MESSAGE = "Error ocurred related to LLM"
+
+
+class ToolEmbeddingError(SdkError):
+    DEFAULT_MESSAGE = "Error ocurred related to embedding"
+
+
+class ToolVectorDBError(SdkError):
+    DEFAULT_MESSAGE = "Error ocurred related to vector DB"
+
+
+class RateLimitError(SdkError):
+    DEFAULT_MESSAGE = "Running into rate limit errors, please try again later"

src/unstract/sdk/exceptions.py

@@ -230,6 +230,9 @@ def index_file(
             }
         )
 
+        if not extracted_text:
+            raise IndexingError("No text available to index")
+
         # Check if chunking is required
         documents = []
         for item in full_text:

src/unstract/sdk/index.py

@@ -1,28 +1,27 @@
 import logging
 import re
-import time
 from typing import Any, Optional
 
 from llama_index.core.llms import LLM, CompletionResponse
+from openai import APIError as OpenAIAPIError
+from openai import RateLimitError as OpenAIRateLimitError
 from unstract.adapters.constants import Common
 from unstract.adapters.llm import adapters
 from unstract.adapters.llm.llm_adapter import LLMAdapter
 
 from unstract.sdk.adapters import ToolAdapter
 from unstract.sdk.constants import LogLevel
-from unstract.sdk.exceptions import SdkError
+from unstract.sdk.exceptions import RateLimitError, SdkError, ToolLLMError
 from unstract.sdk.tool.base import BaseTool
-from unstract.sdk.utils.callback_manager import (
-    CallbackManager as UNCallbackManager,
-)
+from unstract.sdk.utils.callback_manager import CallbackManager as UNCallbackManager
 
 logger = logging.getLogger(__name__)
 
 
 class ToolLLM:
     """Class to handle LLMs for Unstract Tools."""
 
-    json_regex = re.compile(r"\{(?:.|\n)*\}")
+    json_regex = re.compile(r"\[(?:.|\n)*\]|\{(?:.|\n)*\}")
 
     def __init__(self, tool: BaseTool):
         """ToolLLM constructor.
@@ -62,21 +61,21 @@ def run_completion(
             "run_id",
         ]:
             new_kwargs.pop(key, None)
-        for i in range(retries):
-            try:
-                response: CompletionResponse = llm.complete(
-                    prompt, **new_kwargs
-                )
-                match = cls.json_regex.search(response.text)
-                if match:
-                    response.text = match.group(0)
-                return {"response": response}
-
-            except Exception as e:
-                if i == retries - 1:
-                    raise e
-                time.sleep(5)
-        return None
+
+        try:
+            response: CompletionResponse = llm.complete(prompt, **new_kwargs)
+            match = cls.json_regex.search(response.text)
+            if match:
+                response.text = match.group(0)
+            return {"response": response}
+        # TODO: Handle for all LLM providers
+        except OpenAIAPIError as e:
+            msg = e.message
+            if hasattr(e, "body") and "message" in e.body:
+                msg = e.body["message"]
+            if isinstance(e, OpenAIRateLimitError):
+                raise RateLimitError(msg)
+            raise ToolLLMError(msg) from e
 
     def get_llm(self, adapter_instance_id: str) -> LLM:
         """Returns the LLM object for the tool.
@@ -91,13 +90,11 @@ def get_llm(self, adapter_instance_id: str) -> LLM:
             )
             llm_adapter_id = llm_config_data.get(Common.ADAPTER_ID)
             if llm_adapter_id not in self.llm_adapters:
-                raise SdkError(
-                    f"LLM adapter not supported : " f"{llm_adapter_id}"
-                )
+                raise SdkError(f"LLM adapter not supported : " f"{llm_adapter_id}")
 
-            llm_adapter = self.llm_adapters[llm_adapter_id][
-                Common.METADATA
-            ][Common.ADAPTER]
+            llm_adapter = self.llm_adapters[llm_adapter_id][Common.METADATA][
+                Common.ADAPTER
+            ]
             llm_metadata = llm_config_data.get(Common.ADAPTER_METADATA)
             llm_adapter_class: LLMAdapter = llm_adapter(llm_metadata)
             llm_instance: LLM = llm_adapter_class.get_llm_instance()
@@ -106,7 +103,7 @@ def get_llm(self, adapter_instance_id: str) -> LLM:
             self.tool.stream_log(
                 log=f"Unable to get llm instance: {e}", level=LogLevel.ERROR
             )
-            raise SdkError(f"Error getting llm instance: {e}")
+            raise ToolLLMError(f"Error getting llm instance: {e}")
 
     def get_max_tokens(self, reserved_for_output: int = 0) -> int:
         """Returns the maximum number of tokens that can be used for the LLM.

src/unstract/sdk/llm.py

@@ -1,17 +1,14 @@
 import logging
 from typing import Union
 
-from llama_index.core.vector_stores.types import (
-    BasePydanticVectorStore,
-    VectorStore,
-)
+from llama_index.core.vector_stores.types import BasePydanticVectorStore, VectorStore
 from unstract.adapters.constants import Common
 from unstract.adapters.vectordb import adapters
 from unstract.adapters.vectordb.constants import VectorDbConstants
 
 from unstract.sdk.adapters import ToolAdapter
 from unstract.sdk.constants import LogLevel, ToolEnv
-from unstract.sdk.exceptions import SdkError
+from unstract.sdk.exceptions import SdkError, ToolVectorDBError
 from unstract.sdk.platform import PlatformHelper
 from unstract.sdk.tool.base import BaseTool
 
@@ -58,8 +55,7 @@ def get_vector_db(
             vector_db_adapter_id = vector_db_config.get(Common.ADAPTER_ID)
             if vector_db_adapter_id not in self.vector_db_adapters:
                 raise SdkError(
-                    f"VectorDB adapter not supported : "
-                    f"{vector_db_adapter_id}"
+                    f"VectorDB adapter not supported : " f"{vector_db_adapter_id}"
                 )
 
             vector_db_adapter = self.vector_db_adapters[vector_db_adapter_id][
@@ -81,4 +77,4 @@ def get_vector_db(
                 log=f"Unable to get vector_db {adapter_instance_id}: {e}",
                 level=LogLevel.ERROR,
             )
-            raise SdkError(f"Error getting vectorDB instance: {e}")
+            raise ToolVectorDBError(f"Error getting vectorDB instance: {e}")