Fixes LC document deserialization

Depends on apache/burr#175.

Author: skrawcz

scrapegraphai/graphs/smart_scraper_graph_burr.py

@@ -1,7 +1,7 @@
 """
 SmartScraperGraph Module Burr Version
 """
-from typing import Tuple
+from typing import Tuple, Union
 
 from burr import tracking
 from burr.core import Application, ApplicationBuilder, State, default, when
@@ -14,6 +14,7 @@
 from langchain_community.document_transformers import Html2TextTransformer, EmbeddingsRedundantFilter
 from langchain_community.vectorstores import FAISS
 from langchain_core.documents import Document
+from langchain_core import load as lc_serde
 from langchain_core.output_parsers import JsonOutputParser
 from langchain_core.prompts import PromptTemplate
 from langchain_core.runnables import RunnableParallel
@@ -67,10 +68,10 @@ def parse_node(state: State, chunk_size: int = 4096) -> tuple[dict, State]:
 
 @action(reads=["user_prompt", "parsed_doc", "doc"],
         writes=["relevant_chunks"])
-def rag_node(state: State, llm_model: object, embedder_model: object) -> tuple[dict, State]:
-    # bug around input serialization with tracker
-    llm_model = OpenAI({"model_name": "gpt-3.5-turbo"})
-    embedder_model = OpenAIEmbeddings()
+def rag_node(state: State, llm_model: str, embedder_model: object) -> tuple[dict, State]:
+    # bug around input serialization with tracker -- so instantiate objects here:
+    llm_model = OpenAI({"model_name": llm_model})
+    embedder_model = OpenAIEmbeddings() if embedder_model == "openai" else None
     user_prompt = state["user_prompt"]
     doc = state["parsed_doc"]
 
@@ -104,8 +105,10 @@ def rag_node(state: State, llm_model: object, embedder_model: object) -> tuple[d
 
 @action(reads=["user_prompt", "relevant_chunks", "parsed_doc", "doc"],
         writes=["answer"])
-def generate_answer_node(state: State, llm_model: object) -> tuple[dict, State]:
-    llm_model = OpenAI({"model_name": "gpt-3.5-turbo"})
+def generate_answer_node(state: State, llm_model: str) -> tuple[dict, State]:
+    # bug around input serialization with tracker -- so instantiate objects here:
+    llm_model = OpenAI({"model_name": llm_model})
+
     user_prompt = state["user_prompt"]
     doc = state.get("relevant_chunks",
                     state.get("parsed_doc",
@@ -207,21 +210,49 @@ def post_run_step(
     ):
         print(f"Finishing action: {action.name}")
 
+import json
+
+def _deserialize_document(x: Union[str, dict]) -> Document:
+    if isinstance(x, dict):
+        return lc_serde.load(x)
+    elif isinstance(x, str):
+        try:
+            return lc_serde.loads(x)
+        except json.JSONDecodeError:
+            return Document(page_content=x)
+    raise ValueError("Couldn't deserialize document")
+
 
 def run(prompt: str, input_key: str, source: str, config: dict) -> str:
+    # these configs aren't really used yet.
     llm_model = config["llm_model"]
-
     embedder_model = config["embedder_model"]
-    open_ai_embedder = OpenAIEmbeddings()
+    # open_ai_embedder = OpenAIEmbeddings()
     chunk_size = config["model_token"]
 
+    tracker = tracking.LocalTrackingClient(project="smart-scraper-graph")
+    app_instance_id = "testing-12345678919"
     initial_state = {
         "user_prompt": prompt,
         input_key: source,
     }
-    from burr.core import expr
-    tracker = tracking.LocalTrackingClient(project="smart-scraper-graph")
-
+    entry_point = "fetch_node"
+    if app_instance_id:
+        persisted_state = tracker.load(None, app_id=app_instance_id, sequence_no=None)
+        if not persisted_state:
+            print(f"Warning: No persisted state found for app_id {app_instance_id}.")
+        else:
+            initial_state = persisted_state["state"]
+            # for now we need to manually deserialize LangChain messages into LangChain Objects
+            # i.e. we know which objects need to be LC objects
+            initial_state = initial_state.update(**{
+                "doc": _deserialize_document(initial_state["doc"])
+            })
+            docs = [_deserialize_document(doc) for doc in initial_state["relevant_chunks"]]
+            initial_state = initial_state.update(**{
+                "relevant_chunks": docs
+            })
+            entry_point = persisted_state["position"]
 
     app = (
         ApplicationBuilder()
@@ -236,16 +267,17 @@ def run(prompt: str, input_key: str, source: str, config: dict) -> str:
             ("parse_node", "rag_node", default),
             ("rag_node", "generate_answer_node", default)
         )
-        # .with_entrypoint("fetch_node")
-        # .with_state(**initial_state)
-        .initialize_from(
-            tracker,
-            resume_at_next_action=True,  # always resume from entrypoint in the case of failure
-            default_state=initial_state,
-            default_entrypoint="fetch_node",
-        )
-        # .with_identifiers(app_id="testing-123456")
-        .with_tracker(project="smart-scraper-graph")
+        .with_entrypoint(entry_point)
+        .with_state(**initial_state)
+        # this will work once we get serialization plugin for langchain objects done
+        # .initialize_from(
+        #     tracker,
+        #     resume_at_next_action=True,  # always resume from entrypoint in the case of failure
+        #     default_state=initial_state,
+        #     default_entrypoint="fetch_node",
+        # )
+        .with_identifiers(app_id=app_instance_id)
+        .with_tracker(tracker)
         .with_hooks(PrintLnHook())
         .build()
     )
@@ -270,8 +302,8 @@ def run(prompt: str, input_key: str, source: str, config: dict) -> str:
     source = "https://en.wikipedia.org/wiki/Paris"
     input_key = "url"
     config = {
-        "llm_model": "rag-token",
-        "embedder_model": "foo",
+        "llm_model": "gpt-3.5-turbo",
+        "embedder_model": "openai",
         "model_token": "bar",
     }
-    run(prompt, input_key, source, config)
+    print(run(prompt, input_key, source, config))