WEBDIENSTE/langgraph
1
0
Fork 0
mirror of https://github.com/langchain-ai/langgraph.git synced 2026-09-23 18:15:08 +02:00
Code Issues Packages Projects Releases Wiki Activity

Minor updates

Browse Source
This commit is contained in:
Lance Martin
2024-02-07 16:47:32 -08:00
parent addc79d29d
commit 09343a4013
3 changed files with 344 additions and 84 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/rag/langgraph_crag.ipynb
+119 -37
View File
@@ -20,9 +20,23 @@
"id": "8889a307-fa3f-4d38-9127-d41e4686ae47",
"metadata": {},
"source": [
"# CRAG\n",
"# Corrective RAG (CRAG)\n",
"\n",
"Corrective-RAG is a recent paper that introduces an interesting approach for active RAG. \n",
"Self-reflection can enhance RAG, enabling correction of poor quality retrieval or generations.\n",
"\n",
"Several recent papers focus on this theme, but implementing the ideas can be tricky.\n",
"\n",
"Here we show how to implement ideas from the `Corrective RAG (CRAG)` paper [here](https://arxiv.org/pdf/2401.15884.pdf) using LangGraph.\n",
"\n",
"## Dependencies\n",
"\n",
"Set `OPENAI_API_KEY`\n",
"\n",
"Set `TAVILY_API_KEY` to enable web search [here](https://app.tavily.com/sign-in)\n",
"\n",
"## CRAG Detail\n",
"\n",
"Corrective-RAG (CRAG) is a recent paper that introduces an interesting approach for self-reflective RAG. \n",
"\n",
"The framework grades retrieved documents relative to the question:\n",
"\n",
@@ -41,22 +55,9 @@
"\n",
"![Screenshot 2024-02-04 at 2.50.32 PM.png](attachment:5bfa38a2-78a1-4e99-80a2-d98c8a440ea2.png)\n",
"\n",
"Paper -\n",
"\n",
"https://arxiv.org/pdf/2401.15884.pdf\n",
"\n",
"---\n",
"\n",
"Let's implement this from scratch using [LangGraph](https://python.langchain.com/docs/langgraph).\n",
"\n",
"We can make some simplifications:\n",
"\n",
"* Let's skip the knowledge refinement phase as a first pass. This can be added back as a node, if desired. \n",
"* If *any* document is irrelevant, let's opt to supplement retrieval with web search. \n",
"* We'll use [Tavily Search](https://python.langchain.com/docs/integrations/tools/tavily_search) for web search.\n",
"* Let's use query re-writing to optimize the query for web search.\n",
"\n",
"Set the `TAVILY_API_KEY`."
"Let's implement some of these ideas from scratch using [LangGraph](https://python.langchain.com/docs/langgraph)."
]
},
{
@@ -71,7 +72,7 @@
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 4,
"id": "3a566a30-cf0e-4330-ad4d-9bf994bdfa86",
"metadata": {},
"outputs": [],
@@ -120,7 +121,7 @@
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 5,
"id": "94b3945f-ef0f-458d-a443-f763903550b0",
"metadata": {},
"outputs": [],
@@ -157,14 +158,21 @@
"\n",
"Each `edge` will choose which `node` to call next.\n",
"\n",
"It will follow the graph diagram shown above.\n",
"We can make some simplifications from the paper:\n",
"\n",
"* Let's skip the knowledge refinement phase as a first pass. This can be added back as a node, if desired. \n",
"* If *any* document is irrelevant, let's opt to supplement retrieval with web search. \n",
"* We'll use [Tavily Search](https://python.langchain.com/docs/integrations/tools/tavily_search) for web search.\n",
"* Let's use query re-writing to optimize the query for web search.\n",
"\n",
"Here is our graph flow:\n",
"\n",
"![Screenshot 2024-02-04 at 1.32.52 PM.png](attachment:3b65f495-5fc4-497b-83e2-73844a97f6cc.png)"
]
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 7,
"id": "efd639c5-82e2-45e6-a94a-6a4039646ef5",
"metadata": {},
"outputs": [],
@@ -174,7 +182,6 @@
"from typing import Annotated, Sequence, TypedDict\n",
"\n",
"from langchain import hub\n",
"from langchain.output_parsers import PydanticOutputParser\n",
"from langchain.output_parsers.openai_tools import PydanticToolsParser\n",
"from langchain.prompts import PromptTemplate\n",
"from langchain.schema import Document\n",
@@ -186,7 +193,6 @@
"from langchain_core.runnables import RunnablePassthrough\n",
"from langchain_core.utils.function_calling import convert_to_openai_tool\n",
"from langchain_openai import ChatOpenAI, OpenAIEmbeddings\n",
"from langgraph.prebuilt import ToolInvocation\n",
"\n",
"### Nodes ###\n",
"\n",
@@ -415,12 +421,14 @@
"id": "fa076e90-7132-4fcf-8507-db5990314c4f",
"metadata": {},
"source": [
"## Build Graph"
"## Build Graph\n",
"\n",
"The just follows the flow we outlined in the figure above."
]
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 8,
"id": "dedae17a-98c6-474d-90a7-9234b7c8cea0",
"metadata": {},
"outputs": [],
@@ -459,36 +467,110 @@
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 9,
"id": "f5b7c2fe-1fc7-4b76-bf93-ba701a40aa6b",
"metadata": {},
"outputs": [],
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"---RETRIEVE---\n",
"\"Node 'retrieve':\"\n",
"'\\n---\\n'\n",
"---CHECK RELEVANCE---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"\"Node 'grade_documents':\"\n",
"'\\n---\\n'\n",
"---DECIDE TO GENERATE---\n",
"---DECISION: GENERATE---\n",
"---GENERATE---\n",
"\"Node 'generate':\"\n",
"'\\n---\\n'\n",
"\"Node '__end__':\"\n",
"'\\n---\\n'\n",
"('There are several types of memory in human brains, including sensory memory, '\n",
" 'which retains impressions of sensory information for a few seconds after the '\n",
" 'original stimuli have ended. Short-term memory is utilized for in-context '\n",
" 'learning, while long-term memory allows the agent to retain and recall '\n",
" 'information over extended periods by leveraging an external vector store and '\n",
" 'fast retrieval. Additionally, agents can use tool use to call external APIs '\n",
" 'for extra information that is missing from the model weights.')\n"
]
}
],
"source": [
"# Run\n",
"inputs = {\"keys\": {\"question\": \"Explain how the different types of agent memory work?\"}}\n",
"for output in app.stream(inputs):\n",
" for key, value in output.items():\n",
" pprint.pprint(f\"Output from node '{key}':\")\n",
" pprint.pprint(\"---\")\n",
" pprint.pprint(value[\"keys\"], indent=2, width=80, depth=None)\n",
" pprint.pprint(\"\\n---\\n\")"
" # Node\n",
" pprint.pprint(f\"Node '{key}':\")\n",
" # Optional: print full state at each node\n",
" # pprint.pprint(value[\"keys\"], indent=2, width=80, depth=None)\n",
" pprint.pprint(\"\\n---\\n\")\n",
"\n",
"# Final generation\n",
"pprint.pprint(value['keys']['generation'])"
]
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 11,
"id": "2bee03de-a32c-4bbe-b37a-a13bb825e4cb",
"metadata": {},
"outputs": [],
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"---RETRIEVE---\n",
"\"Node 'retrieve':\"\n",
"'\\n---\\n'\n",
"---CHECK RELEVANCE---\n",
"---GRADE: DOCUMENT NOT RELEVANT---\n",
"---GRADE: DOCUMENT NOT RELEVANT---\n",
"---GRADE: DOCUMENT NOT RELEVANT---\n",
"---GRADE: DOCUMENT NOT RELEVANT---\n",
"\"Node 'grade_documents':\"\n",
"'\\n---\\n'\n",
"---DECIDE TO GENERATE---\n",
"---DECISION: TRANSFORM QUERY and RUN WEB SEARCH---\n",
"---TRANSFORM QUERY---\n",
"\"Node 'transform_query':\"\n",
"'\\n---\\n'\n",
"---WEB SEARCH---\n",
"\"Node 'web_search':\"\n",
"'\\n---\\n'\n",
"---GENERATE---\n",
"\"Node 'generate':\"\n",
"'\\n---\\n'\n",
"\"Node '__end__':\"\n",
"'\\n---\\n'\n",
"('The AlphaCodium paper uses a test-based, iterative approach for code '\n",
" 'generation. It employs a multi-stage, code-oriented flow that addresses the '\n",
" 'specific challenges of coding problems. Unlike traditional models, '\n",
" 'AlphaCodium actively engages in problem self-reflection, reasoning, and '\n",
" 'iterative code solution generation.')\n"
]
}
],
"source": [
"# Correction for question not present in context\n",
"inputs = {\"keys\": {\"question\": \"What is the approach taken in the AlphaCodium paper?\"}}\n",
"inputs = {\"keys\": {\"question\": \"What is the approach for code generation taken in the AlphaCodium paper?\"}}\n",
"for output in app.stream(inputs):\n",
" for key, value in output.items():\n",
" pprint.pprint(f\"Output from node '{key}':\")\n",
" pprint.pprint(\"---\")\n",
" pprint.pprint(value[\"keys\"], indent=2, width=80, depth=None)\n",
" pprint.pprint(\"\\n---\\n\")"
" # Node\n",
" pprint.pprint(f\"Node '{key}':\")\n",
" # Optional: print full state \n",
" # pprint.pprint(value[\"keys\"], indent=2, width=80, depth=None)\n",
" pprint.pprint(\"\\n---\\n\")\n",
"\n",
"# Final generation\n",
"pprint.pprint(value['keys']['generation'])"
]
},
{
examples/rag/langgraph_crag_mistral.ipynb
+105 -20
View File
@@ -29,7 +29,7 @@
"id": "92ddc4f4-f7bf-4e0e-b5a5-5abd8a008b21",
"metadata": {},
"source": [
"# Self-Reflective RAG\n",
"# Corrective RAG\n",
"\n",
"Self-reflection can enhance RAG, enabling correction of poor quality retrieval or generations.\n",
"\n",
@@ -37,7 +37,7 @@
"\n",
"Here we show how to implement self-reflective RAG using `Mistral` and `LangGraph`.\n",
"\n",
"In particular, we'll focus on the approach from one paper focused on Corrective RAG (CRAG) [here](https://arxiv.org/pdf/2401.15884.pdf).\n",
"We'll focus on ideas from one paper, `Corrective RAG (CRAG)` [here](https://arxiv.org/pdf/2401.15884.pdf).\n",
"\n",
"![Screenshot 2024-02-07 at 1.21.51 PM.png](attachment:9db7f9db-55aa-48cb-95d5-bcde3f937589.png)\n",
"\n",
@@ -187,7 +187,7 @@
},
{
"cell_type": "code",
"execution_count": 53,
"execution_count": 5,
"id": "447d1333-082d-479a-a6fa-0ac0df78bb9d",
"metadata": {},
"outputs": [],
@@ -461,12 +461,14 @@
"id": "6096626d-dfa5-48e0-8a24-3747b298bc67",
"metadata": {},
"source": [
"## Build Graph"
"## Build Graph\n",
"\n",
"The just follows the flow we outlined in the figure above."
]
},
{
"cell_type": "code",
"execution_count": 54,
"execution_count": 6,
"id": "0a63776c-f9cd-46ce-b8cf-95c066dc5b06",
"metadata": {},
"outputs": [],
@@ -515,12 +517,48 @@
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 13,
"id": "3ab1d8df-a74e-4b48-a30b-e39bbfd5925a",
"metadata": {
"scrolled": true
},
"outputs": [],
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"---RETRIEVE---\n",
"\"Node 'retrieve':\"\n",
"'\\n---\\n'\n",
"---CHECK RELEVANCE---\n",
"---GRADE: DOCUMENT NOT RELEVANT---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"\"Node 'grade_documents':\"\n",
"'\\n---\\n'\n",
"---DECIDE TO GENERATE---\n",
"---DECISION: TRANSFORM QUERY and RUN WEB SEARCH---\n",
"---TRANSFORM QUERY---\n",
"\"Node 'transform_query':\"\n",
"'\\n---\\n'\n",
"---WEB SEARCH---\n",
"\"Node 'web_search':\"\n",
"'\\n---\\n'\n",
"---GENERATE---\n",
"\"Node 'generate':\"\n",
"'\\n---\\n'\n",
"\"Node '__end__':\"\n",
"'\\n---\\n'\n",
"('Episodic memory stores specific events or experiences, making them unique to '\n",
" 'each individual. Semantic memory, on the other hand, involves general '\n",
" 'knowledge and facts that are not tied to personal experiences. Procedural '\n",
" 'memory is responsible for learning and remembering sequences of actions, '\n",
" \"such as riding a bike. These memory types contribute to an agent's learning \"\n",
" 'and decision-making processes by allowing it to recall past experiences '\n",
" '(episodic), understand and use information (semantic), and perform tasks '\n",
" '(procedural).')\n"
]
}
],
"source": [
"# Run\n",
"inputs = {\n",
@@ -531,10 +569,14 @@
"}\n",
"for output in app.stream(inputs):\n",
" for key, value in output.items():\n",
" pprint.pprint(f\"Output from node '{key}':\")\n",
" pprint.pprint(\"---\")\n",
" pprint.pprint(value[\"keys\"], indent=2, width=80, depth=None)\n",
" pprint.pprint(\"\\n---\\n\")"
" # Node\n",
" pprint.pprint(f\"Node '{key}':\")\n",
" # Optional: print full state at each node\n",
" # pprint.pprint(value[\"keys\"], indent=2, width=80, depth=None)\n",
" pprint.pprint(\"\\n---\\n\")\n",
"\n",
"# Final generation\n",
"pprint.pprint(value['keys']['generation'])"
]
},
{
@@ -547,10 +589,49 @@
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 14,
"id": "16ea2032-59c7-433d-aca4-2828a1239074",
"metadata": {},
"outputs": [],
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"---RETRIEVE---\n",
"\"Node 'retrieve':\"\n",
"'\\n---\\n'\n",
"---CHECK RELEVANCE---\n",
"---GRADE: DOCUMENT NOT RELEVANT---\n",
"---GRADE: DOCUMENT NOT RELEVANT---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"\"Node 'grade_documents':\"\n",
"'\\n---\\n'\n",
"---DECIDE TO GENERATE---\n",
"---DECISION: TRANSFORM QUERY and RUN WEB SEARCH---\n",
"---TRANSFORM QUERY---\n",
"\"Node 'transform_query':\"\n",
"'\\n---\\n'\n",
"---WEB SEARCH---\n",
"\"Node 'web_search':\"\n",
"'\\n---\\n'\n",
"---GENERATE---\n",
"\"Node 'generate':\"\n",
"'\\n---\\n'\n",
"\"Node '__end__':\"\n",
"'\\n---\\n'\n",
"('There are three types of agent memory in artificial intelligence systems: '\n",
" 'sensory memory, short-term memory, and long-term memory. Sensory memory is '\n",
" 'the learning embedding representations for raw inputs such as text, image or '\n",
" 'other modalities. Short-term memory is in-context learning that is short and '\n",
" 'finite, restricted by the finite context window length of Transformer. '\n",
" 'Long-term memory is an external vector store that the agent can attend to at '\n",
" 'query time, accessible via fast retrieval. The external memory can alleviate '\n",
" 'the restriction of finite attention span by using approximate nearest '\n",
" 'neighbors (ANN) algorithms such as maximum inner product search (MIPS).')\n"
]
}
],
"source": [
"# Run\n",
"inputs = {\n",
@@ -561,10 +642,14 @@
"}\n",
"for output in app.stream(inputs):\n",
" for key, value in output.items():\n",
" pprint.pprint(f\"Output from node '{key}':\")\n",
" pprint.pprint(\"---\")\n",
" pprint.pprint(value[\"keys\"], indent=2, width=80, depth=None)\n",
" pprint.pprint(\"\\n---\\n\")"
" # Node\n",
" pprint.pprint(f\"Node '{key}':\")\n",
" # Optional: print full state at each node\n",
" # pprint.pprint(value[\"keys\"], indent=2, width=80, depth=None)\n",
" pprint.pprint(\"\\n---\\n\")\n",
"\n",
"# Final generation\n",
"pprint.pprint(value['keys']['generation'])"
]
},
{
examples/rag/langgraph_self_rag.ipynb
+120 -27
View File
@@ -22,9 +22,21 @@
"source": [
"# Self-RAG\n",
"\n",
"Self-RAG is a recent paper that introduces an interesting approach for active RAG. \n",
"Self-reflection can enhance RAG, enabling correction of poor quality retrieval or generations.\n",
"\n",
"The framework trains a single arbitrary LM (LLaMA2-7b, 13b) to generate tokens that govern the RAG process:\n",
"Several recent papers focus on this theme, but implementing the ideas can be tricky.\n",
"\n",
"Here we show how to implement ideas from the `Self RAG` paper [here](https://arxiv.org/abs/2310.11511) using LangGraph.\n",
"\n",
"## Dependencies\n",
"\n",
"Set `OPENAI_API_KEY`\n",
"\n",
"## Self-RAG Detail\n",
"\n",
"Self-RAG is a recent paper that introduces an interesting approach for self-reflective RAG. \n",
"\n",
"The framework trains an LLM (e.g., LLaMA2-7b or 13b) to generate tokens that govern the RAG process in a few ways:\n",
"\n",
"1. Should I retrieve from retriever, `R` -\n",
"\n",
@@ -59,13 +71,9 @@
"\n",
"![Screenshot 2024-02-02 at 1.36.44 PM.png](attachment:ea6a57d2-f2ec-4061-840a-98deb3207248.png)\n",
"\n",
"Paper -\n",
"\n",
"https://arxiv.org/abs/2310.11511\n",
"\n",
"---\n",
"\n",
"Let's implement this from scratch using [LangGraph](https://python.langchain.com/docs/langgraph)."
"Let's implement some of these ideas from scratch using [LangGraph](https://python.langchain.com/docs/langgraph)."
]
},
{
@@ -80,7 +88,7 @@
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 1,
"id": "565a6d44-2c9f-4fff-b1ec-eea05df9350d",
"metadata": {},
"outputs": [],
@@ -129,7 +137,7 @@
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 2,
"id": "f1617e9e-66a8-4c1a-a1fe-cc936284c085",
"metadata": {},
"outputs": [],
@@ -166,14 +174,16 @@
"\n",
"Each `edge` will choose which `node` to call next.\n",
"\n",
"We can lay out `self-RAG` as a graph:\n",
"We can lay out `self-RAG` as a graph.\n",
"\n",
"Here is our graph flow:\n",
"\n",
"![Screenshot 2024-02-02 at 9.01.01 PM.png](attachment:e61fbd0c-e667-4160-a96c-82f95a560b44.png)"
]
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 3,
"id": "add509d8-6682-4127-8d95-13dd37d79702",
"metadata": {},
"outputs": [],
@@ -183,7 +193,6 @@
"from typing import Annotated, Sequence, TypedDict\n",
"\n",
"from langchain import hub\n",
"from langchain.output_parsers import PydanticOutputParser\n",
"from langchain.output_parsers.openai_tools import PydanticToolsParser\n",
"from langchain.prompts import PromptTemplate\n",
"from langchain_community.vectorstores import Chroma\n",
@@ -193,7 +202,6 @@
"from langchain_core.runnables import RunnablePassthrough\n",
"from langchain_core.utils.function_calling import convert_to_openai_tool\n",
"from langchain_openai import ChatOpenAI, OpenAIEmbeddings\n",
"from langgraph.prebuilt import ToolInvocation\n",
"\n",
"### Nodes ###\n",
"\n",
@@ -538,12 +546,14 @@
"id": "61cd5797-1782-4d78-a277-8196d13f3e1b",
"metadata": {},
"source": [
"## Build Graph"
"## Build Graph\n",
"\n",
"The just follows the flow we outlined in the figure above."
]
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 5,
"id": "0e09ca9f-e36d-4ef4-a0d5-79fdbada9fe0",
"metadata": {},
"outputs": [],
@@ -596,35 +606,118 @@
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 6,
"id": "fb69dbb9-91ee-4868-8c3c-93af3cd885be",
"metadata": {},
"outputs": [],
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"---RETRIEVE---\n",
"\"Node 'retrieve':\"\n",
"'\\n---\\n'\n",
"---CHECK RELEVANCE---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"\"Node 'grade_documents':\"\n",
"'\\n---\\n'\n",
"---DECIDE TO GENERATE---\n",
"---DECISION: GENERATE---\n",
"---GENERATE---\n",
"\"Node 'generate':\"\n",
"'\\n---\\n'\n",
"---GRADE GENERATION vs DOCUMENTS---\n",
"---DECISION: SUPPORTED, MOVE TO FINAL GRADE---\n",
"---FINAL GRADE---\n",
"\"Node 'prepare_for_final_grade':\"\n",
"'\\n---\\n'\n",
"---GRADE GENERATION vs QUESTION---\n",
"---DECISION: USEFUL---\n",
"\"Node '__end__':\"\n",
"'\\n---\\n'\n",
"('Short-term memory is the stage of memory that stores information that we are '\n",
" 'currently aware of and needed to carry out complex cognitive tasks. It has a '\n",
" 'limited capacity and lasts for a short duration. Long-term memory, on the '\n",
" 'other hand, can store information for a long time and has unlimited storage '\n",
" 'capacity. It includes explicit/declarative memory for facts and events, and '\n",
" 'implicit/procedural memory for unconscious skills and routines.')\n"
]
}
],
"source": [
"# Run\n",
"inputs = {\"keys\": {\"question\": \"Explain how the different types of agent memory work?\"}}\n",
"for output in app.stream(inputs):\n",
" for key, value in output.items():\n",
" pprint.pprint(f\"Output from node '{key}':\")\n",
" pprint.pprint(\"---\")\n",
" pprint.pprint(value[\"keys\"], indent=2, width=80, depth=None)\n",
" pprint.pprint(\"\\n---\\n\")"
" # Node\n",
" pprint.pprint(f\"Node '{key}':\")\n",
" # Optional: print full state at each node\n",
" # pprint.pprint(value[\"keys\"], indent=2, width=80, depth=None)\n",
" pprint.pprint(\"\\n---\\n\")\n",
"\n",
"# Final generation\n",
"pprint.pprint(value['keys']['generation'])"
]
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 7,
"id": "4138bc51-8c84-4b8a-8d24-f7f470721f6f",
"metadata": {},
"outputs": [],
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"---RETRIEVE---\n",
"\"Node 'retrieve':\"\n",
"'\\n---\\n'\n",
"---CHECK RELEVANCE---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"---GRADE: DOCUMENT RELEVANT---\n",
"\"Node 'grade_documents':\"\n",
"'\\n---\\n'\n",
"---DECIDE TO GENERATE---\n",
"---DECISION: GENERATE---\n",
"---GENERATE---\n",
"\"Node 'generate':\"\n",
"'\\n---\\n'\n",
"---GRADE GENERATION vs DOCUMENTS---\n",
"---DECISION: SUPPORTED, MOVE TO FINAL GRADE---\n",
"---FINAL GRADE---\n",
"\"Node 'prepare_for_final_grade':\"\n",
"'\\n---\\n'\n",
"---GRADE GENERATION vs QUESTION---\n",
"---DECISION: USEFUL---\n",
"\"Node '__end__':\"\n",
"'\\n---\\n'\n",
"('Chain of thought prompting involves guiding the behavior of autoregressive '\n",
" 'language models by providing prompts or demonstrations that contain '\n",
" 'high-quality reasoning chains. This can be done through methods such as '\n",
" 'self-asking, interleaving retrieval with chain-of-thought reasoning, and '\n",
" 'complexity-based prompting for multi-step reasoning. These techniques aim to '\n",
" \"improve the model's ability to generate coherent and logical responses \"\n",
" 'without updating its weights.')\n"
]
}
],
"source": [
"inputs = {\"keys\": {\"question\": \"Explain how chain of thought prompting works?\"}}\n",
"for output in app.stream(inputs):\n",
" for key, value in output.items():\n",
" pprint.pprint(f\"Output from node '{key}':\")\n",
" pprint.pprint(\"---\")\n",
" pprint.pprint(value[\"keys\"], indent=2, width=80, depth=None)\n",
" pprint.pprint(\"\\n---\\n\")"
" # Node\n",
" pprint.pprint(f\"Node '{key}':\")\n",
" # Optional: print full state at each node\n",
" # pprint.pprint(value[\"keys\"], indent=2, width=80, depth=None)\n",
" pprint.pprint(\"\\n---\\n\")\n",
"\n",
"# Final generation\n",
"pprint.pprint(value['keys']['generation'])"
]
},
{