github/langchain
1
0
Fork 0
mirror of https://github.com/hwchase17/langchain.git synced 2025-07-18 18:53:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

Minor template cleaning (#12573)

Browse Source
This commit is contained in:
Lance Martin 2023-10-30 11:27:44 -07:00 committed by GitHub
parent b1e3843931
commit 08103e6d48
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
5 changed files with 28 additions and 140 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
templates/llama2-functions/README.md
View File

@ -2,14 +2,17 @@
This template shows how to do extraction of structured data from unstructured data, using LLaMA2 [fine-tuned for grammars and jsonschema](https://replicate.com/andreasjansson/llama-2-13b-chat-gguf).
Specify the scehma you want to extract in `chain.py`
[Query transformations](https://blog.langchain.dev/query-transformations/) are one great application area for open source, private LLMs:
By default, it will extract the title and author of papers.
* The tasks are often narrow and well-defined (e.g., generatae multiple questions from a user input)
* They also are tasks that users may want to run locally (e.g., in a RAG workflow)
Specify the scehma you want to extract in `chain.py`
## LLM
This template will use a `Replicate` [hosted version](https://replicate.com/andreasjansson/llama-2-13b-chat-gguf) of LLaMA2 that has support for grammars and jsonschema.
Based on the `Replicate` example, these are supplied directly in the prompt.
Based on the `Replicate` example, the JSON schema is supplied directly in the prompt.
Be sure that `REPLICATE_API_TOKEN` is set in your environment.

29
templates/llama2-functions/llama2-functions.ipynb
View File

@ -5,30 +5,6 @@
"cell_type": "markdown",
"id": "9faf648c-541e-4368-82a8-96287dbf34de",
"metadata": {},
"source": [
"## Document Loading\n",
"\n",
"Load a blog post on agents."
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "662a843a-49e8-40ec-bd32-0f44bc4159a1",
"metadata": {},
"outputs": [],
"source": [
"from langchain.document_loaders import WebBaseLoader\n",
"\n",
"loader = WebBaseLoader(\"https://lilianweng.github.io/posts/2023-06-23-agent/\")\n",
"text = loader.load()"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "67306dbd-d79c-4723-825e-7d88edb811ba",
"metadata": {},
"source": [
"## Run Template\n",
"\n",
@ -41,14 +17,13 @@
{
"cell_type": "code",
"execution_count": null,
"id": "3668ba4b-254e-4a3b-bfb5-53242572cb1b",
"id": "2dfe28bb-6112-459b-a77d-013964b65409",
"metadata": {},
"outputs": [],
"source": [
"from langserve.client import RemoteRunnable\n",
"\n",
"llama2_function = RemoteRunnable(\"http://0.0.0.0:8001/llama2_functions\")\n",
"llama2_function.invoke({\"input\": text[0].page_content[0:1500]})"
"llama2_function.invoke({\"question\":\"How does agent memory work?\"})"
]
}
],

30
templates/llama2-functions/llama2_functions/chain.py
View File

@ -9,32 +9,34 @@ model = Replicate(
)
# Prompt with output schema specification
template = """A article will be passed to you. Extract from it all papers that are mentioned by this article.
Do not extract the name of the article itself. If no papers are mentioned that's fine - you don't need to extract any! Just return an empty list.
Do not make up or guess ANY extra information. Only extract what exactly is in the text.
Respond with json that adheres to the following jsonschema:
template = """You are an AI language model assistant. Your task is to generate 3 different versions of the given user /
question to retrieve relevant documents from a vector database. By generating multiple perspectives on the user /
question, your goal is to help the user overcome some of the limitations of distance-based similarity search. /
Respond with json that adheres to the following jsonschema:
{{
"$schema": "http://json-schema.org/draft-07/schema#",
"type": "object",
"properties": {{
"author": {{
"question_1": {{
"type": "string",
"description": "The author of the paper."
"description": "First version of the user question."
}},
"title": {{
"question_2": {{
"type": "string",
"description": "The title of the paper."
"description": "Second version of the user question."
}},
"question_3": {{
"type": "string",
"description": "Third version of the user question."
}}
}},
"required": ["author", "title"],
"required": ["question_1","question_2","question_3"],
"additionalProperties": false
}}""" # noqa: E501
prompt = ChatPromptTemplate.from_messages([("system", template), ("human", "{input}")])
prompt = ChatPromptTemplate.from_messages(
[("system", template), ("human", "{question}")]
)
# Chain
chain = prompt | model

97
templates/rag-chroma-private/rag_chroma_private.ipynb
View File

@ -13,20 +13,6 @@
"```"
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "78fb41d3-d2aa-40a6-b144-491f38a7cf88",
"metadata": {},
"outputs": [],
"source": [
"from langserve.client import RemoteRunnable\n",
"\n",
"rag_app = RemoteRunnable(\"http://0.0.0.0:8001/rag_chroma_private/\")\n",
"for item in rag_app.stream(\"How does agent memory work?\"):\n",
" print(item)"
]
},
{
"cell_type": "code",
"execution_count": 3,
@ -41,88 +27,11 @@
]
}
],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "a554971a-e724-4c99-84d1-5d646ae4ac3e",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 2,
"id": "6891d028-43ac-4a70-b2ad-6fbd3d937283",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<generator object RemoteRunnable.stream at 0x1245d25f0>"
]
},
"execution_count": 2,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"rag_app.stream(\"How does agent memory work?\")"
"from langserve.client import RemoteRunnable\n",
"rag_app = RemoteRunnable(\"http://0.0.0.0:8001/rag_chroma_private/\")\n",
"rag_app.invoke(\"How does agent memory work?\")"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "888494ca-0509-4070-b36f-600a042f352c",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
" Based on the provided context, agent memory is a long-term memory module that records a comprehensive list of agents' experiences in natural language. Each element is an observation, an event directly provided by the agent, and inter-agent communication can trigger new natural language statements. The memory module surfaces the context to inform the agent's behavior according to relevance, recency, and importance.\n"
]
}
],
"source": [
"\n",
"stream = \n",
"for i in stream:\n",
" print(i)"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "ff2169c9-dab2-41c4-8f38-1f8aebb16814",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Collecting httpx_sse\n",
" Obtaining dependency information for httpx_sse from https://files.pythonhosted.org/packages/62/33/d35b4ccf8c1ac7266bd1d068c48f842d3c7392cca87e32751c79ee553d7a/httpx_sse-0.3.1-py3-none-any.whl.metadata\n",
" Using cached httpx_sse-0.3.1-py3-none-any.whl.metadata (8.6 kB)\n",
"Using cached httpx_sse-0.3.1-py3-none-any.whl (7.7 kB)\n",
"Installing collected packages: httpx_sse\n",
"Successfully installed httpx_sse-0.3.1\n"
]
}
],
"source": [
"! pip install httpx_sse"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3d843f23-686a-4138-8a9d-087bb00b2e13",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {

3
templates/rag-chroma-private/rag_chroma_private/chain.py
View File

@ -1,6 +1,5 @@
from langchain.chat_models import ChatOllama
# Load
from langchain.chat_models import ChatOllama
from langchain.document_loaders import WebBaseLoader
from langchain.embeddings import GPT4AllEmbeddings
from langchain.prompts import ChatPromptTemplate