fix: actually callable_chain does not work for langchain so we have to make runnables without decorators

Signed-off-by: thiswillbeyourgithub <[email protected]>

Author: thiswillbeyourgithub

wdoc/utils/customs/callable_runnable.py

@@ -42,7 +42,6 @@
 from platformdirs import user_cache_dir
 from loguru import logger
 
-from wdoc.utils.customs.callable_runnable import callable_chain
 from wdoc.utils.env import env, is_input_piped, pytest_ongoing
 from wdoc.utils.errors import UnexpectedDocDictArgument
 
@@ -425,8 +424,6 @@ def html_to_text(html: str, remove_image: bool = False) -> str:
     return text
 
 
-@callable_chain
-@chain
 def debug_chain(inputs: Union[dict, List]) -> Union[dict, List]:
     "use it between | pipes | in a chain to open the debugger"
     if hasattr(inputs, "keys"):
@@ -435,6 +432,9 @@ def debug_chain(inputs: Union[dict, List]) -> Union[dict, List]:
     return inputs
 
 
+debug_chain = chain(debug_chain)
+
+
 def wrapped_model_name_matcher(model: str) -> str:
     "find the best match for a modelname (wrapped to make some check)"
     # find the currently set api keys to avoid matching models from

wdoc/utils/misc.py

@@ -22,7 +22,6 @@
 from tqdm import tqdm
 from loguru import logger
 
-from wdoc.utils.customs.callable_runnable import callable_chain
 from wdoc.utils.env import env
 from wdoc.utils.errors import (
     InvalidDocEvaluationByLLMEval,
@@ -52,8 +51,6 @@ def sieve_documents(instance) -> RunnableLambda:
     we can end up with a lot more document!
     """
 
-    @callable_chain
-    @chain
     def _sieve(inputs: dict) -> dict:
         assert "question_to_answer" in inputs, inputs.keys()
         assert "unfiltered_docs" in inputs, inputs.keys()
@@ -71,11 +68,11 @@ def _sieve(inputs: dict) -> dict:
         inputs["unfiltered_docs"] = inputs["unfiltered_docs"][: instance.top_k]
         return inputs
 
+    _sieve = chain(_sieve)
+
     return _sieve
 
 
-@callable_chain
-@chain
 @log_and_time_fn
 def refilter_docs(inputs: dict) -> List[Document]:
     "filter documents fond via RAG based on the digit answered by the eval llm"
@@ -118,6 +115,9 @@ def refilter_docs(inputs: dict) -> List[Document]:
     return filtered_docs
 
 
+refilter_docs = chain(refilter_docs)
+
+
 @log_and_time_fn
 def parse_eval_output(output: str) -> str:
     """
@@ -536,8 +536,6 @@ def pbar_chain(
 ) -> RunnableLambda:
     "create a chain that just sets a tqdm progress bar"
 
-    @callable_chain
-    @chain
     def actual_pbar_chain(
         inputs: Union[dict, List],
         llm: Union[ChatLiteLLM, FakeListChatModel] = llm,
@@ -554,6 +552,8 @@ def actual_pbar_chain(
 
         return inputs
 
+    actual_pbar_chain = chain(actual_pbar_chain)
+
     return actual_pbar_chain
 
 
@@ -562,8 +562,6 @@ def pbar_closer(
 ) -> RunnableLambda:
     "close a pbar created by pbar_chain"
 
-    @callable_chain
-    @chain
     def actual_pbar_closer(
         inputs: Union[dict, List],
         llm: Union[ChatLiteLLM, FakeListChatModel] = llm,
@@ -575,4 +573,6 @@ def actual_pbar_closer(
 
         return inputs
 
+    actual_pbar_closer = chain(actual_pbar_closer)
+
     return actual_pbar_closer

wdoc/utils/tasks/query.py

@@ -38,7 +38,6 @@
 from tqdm import tqdm
 from loguru import logger as logger
 
-from wdoc.utils.customs.callable_runnable import callable_chain
 from wdoc.utils.batch_file_loader import batch_load_doc
 from wdoc.utils.customs.fix_llm_caching import SQLiteCacheFixed
 from wdoc.utils.embeddings import create_embeddings, load_embeddings_engine
@@ -1415,8 +1414,6 @@ def eval_cache_wrapper(func: Callable) -> Callable:
                 f"Related github issue: 'https://github.com/langchain-ai/langchain/issues/23257'"
             )
 
-        @callable_chain
-        @chain
         def autoincrease_top_k(filtered_docs: List[Document]) -> List[Document]:
             if not self.max_top_k:
                 return filtered_docs
@@ -1439,8 +1436,8 @@ def autoincrease_top_k(filtered_docs: List[Document]) -> List[Document]:
                     )
             return filtered_docs
 
-        @callable_chain
-        @chain
+        autoincrease_top_k = chain(autoincrease_top_k)
+
         @eval_cache_wrapper
         def evaluate_doc_chain(
             inputs: dict,
@@ -1564,6 +1561,8 @@ async def do_eval(subinputs):
                 self.eval_llm.callbacks[0].pbar[-1].update(1)
             return outputs
 
+        evaluate_doc_chain = chain(evaluate_doc_chain)
+
         # uses in most places to increase concurrency limit
         multi = {
             "max_concurrency": env.WDOC_LLM_MAX_CONCURRENCY if not self.debug else 1
@@ -1572,8 +1571,6 @@ async def do_eval(subinputs):
         if self.task == "search":
             if self.query_eval_model is not None:
                 # for some reason I needed to have at least one chain object otherwise rag_chain is a dict
-                @callable_chain
-                @chain
                 def retrieve_documents(inputs):
                     return {
                         "unfiltered_docs": retriever.invoke(
@@ -1583,6 +1580,8 @@ def retrieve_documents(inputs):
                     }
                     return inputs
 
+                retrieve_documents = chain(retrieve_documents)
+
                 meta_refilter_docs = {
                     "filtered_docs": (
                         RunnablePassthrough.assign(
@@ -1774,8 +1773,6 @@ def retrieve_documents(inputs):
 
         else:
             # for some reason I needed to have at least one chain object otherwise rag_chain is a dict
-            @callable_chain
-            @chain
             def retrieve_documents(inputs):
                 return {
                     "unfiltered_docs": retriever.invoke(
@@ -1785,6 +1782,8 @@ def retrieve_documents(inputs):
                 }
                 return inputs
 
+            retrieve_documents = chain(retrieve_documents)
+
             meta_refilter_docs = {
                 "filtered_docs": (
                     RunnablePassthrough.assign(