mirror of
https://github.com/hwchase17/langchain
synced 2024-11-06 03:20:49 +00:00
Update Modal.com integration docs (#8014)
Hey, I'm a Modal Labs engineer and I'm making this docs update after getting a user question in [our beta Slack space](https://join.slack.com/t/modalbetatesters/shared_invite/zt-1xl9gbob8-1QDgUY7_PRPg6dQ49hqEeQ) about the Langchain integration docs. 🔗 [Modal beta-testers link to docs discussion thread](https://modalbetatesters.slack.com/archives/C031Z7DBQFL/p1689777700594819?thread_ts=1689775859.855849&cid=C031Z7DBQFL)
This commit is contained in:
Jonathon Belotti
GitHub
parent
62d0475c29
commit
021bb9be84
@ -1,66 +1,95 @@
|
||||
# Modal
|
||||
|
||||
This page covers how to use the Modal ecosystem within LangChain.
|
||||
It is broken into two parts: installation and setup, and then references to specific Modal wrappers.
|
||||
This page covers how to use the Modal ecosystem to run LangChain custom LLMs.
|
||||
It is broken into two parts:
|
||||
|
||||
1. Modal installation and web endpoint deployment
|
||||
2. Using deployed web endpoint with `LLM` wrapper class.
|
||||
|
||||
## Installation and Setup
|
||||
- Install with `pip install modal-client`
|
||||
|
||||
- Install with `pip install modal`
|
||||
- Run `modal token new`
|
||||
|
||||
## Define your Modal Functions and Webhooks
|
||||
|
||||
You must include a prompt. There is a rigid response structure.
|
||||
You must include a prompt. There is a rigid response structure:
|
||||
|
||||
```python
|
||||
class Item(BaseModel):
|
||||
prompt: str
|
||||
|
||||
@stub.webhook(method="POST")
|
||||
def my_webhook(item: Item):
|
||||
return {"prompt": my_function.call(item.prompt)}
|
||||
@stub.function()
|
||||
@modal.web_endpoint(method="POST")
|
||||
def get_text(item: Item):
|
||||
return {"prompt": run_gpt2.call(item.prompt)}
|
||||
```
|
||||
|
||||
An example with GPT2:
|
||||
The following is an example with the GPT2 model:
|
||||
|
||||
```python
|
||||
from pydantic import BaseModel
|
||||
|
||||
import modal
|
||||
|
||||
stub = modal.Stub("example-get-started")
|
||||
|
||||
volume = modal.SharedVolume().persist("gpt2_model_vol")
|
||||
CACHE_PATH = "/root/model_cache"
|
||||
|
||||
@stub.function(
|
||||
gpu="any",
|
||||
image=modal.Image.debian_slim().pip_install(
|
||||
"tokenizers", "transformers", "torch", "accelerate"
|
||||
),
|
||||
shared_volumes={CACHE_PATH: volume},
|
||||
retries=3,
|
||||
)
|
||||
def run_gpt2(text: str):
|
||||
from transformers import GPT2Tokenizer, GPT2LMHeadModel
|
||||
tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
|
||||
model = GPT2LMHeadModel.from_pretrained('gpt2')
|
||||
encoded_input = tokenizer(text, return_tensors='pt').input_ids
|
||||
output = model.generate(encoded_input, max_length=50, do_sample=True)
|
||||
return tokenizer.decode(output[0], skip_special_tokens=True)
|
||||
|
||||
class Item(BaseModel):
|
||||
prompt: str
|
||||
|
||||
@stub.webhook(method="POST")
|
||||
stub = modal.Stub(name="example-get-started-with-langchain")
|
||||
|
||||
def download_model():
|
||||
from transformers import GPT2Tokenizer, GPT2LMHeadModel
|
||||
tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
|
||||
model = GPT2LMHeadModel.from_pretrained('gpt2')
|
||||
tokenizer.save_pretrained(CACHE_PATH)
|
||||
model.save_pretrained(CACHE_PATH)
|
||||
|
||||
# Define a container image for the LLM function below, which
|
||||
# downloads and stores the GPT-2 model.
|
||||
image = modal.Image.debian_slim().pip_install(
|
||||
"tokenizers", "transformers", "torch", "accelerate"
|
||||
).run_function(download_model)
|
||||
|
||||
@stub.function(
|
||||
gpu="any",
|
||||
image=image,
|
||||
retries=3,
|
||||
)
|
||||
def run_gpt2(text: str):
|
||||
from transformers import GPT2Tokenizer, GPT2LMHeadModel
|
||||
tokenizer = GPT2Tokenizer.from_pretrained(CACHE_PATH)
|
||||
model = GPT2LMHeadModel.from_pretrained(CACHE_PATH)
|
||||
encoded_input = tokenizer(text, return_tensors='pt').input_ids
|
||||
output = model.generate(encoded_input, max_length=50, do_sample=True)
|
||||
return tokenizer.decode(output[0], skip_special_tokens=True)
|
||||
|
||||
@stub.function()
|
||||
@modal.web_endpoint(method="POST")
|
||||
def get_text(item: Item):
|
||||
return {"prompt": run_gpt2.call(item.prompt)}
|
||||
```
|
||||
|
||||
## Wrappers
|
||||
### Deploy the web endpoint
|
||||
|
||||
### LLM
|
||||
Deploy the web endpoint to Modal cloud with the [`modal deploy`](https://modal.com/docs/reference/cli/deploy) CLI command.
|
||||
Your web endpoint will acquire a persistent URL under the `modal.run` domain.
|
||||
|
||||
## LLM wrapper around Modal web endpoint
|
||||
|
||||
The `Modal` LLM wrapper class which will accept your deployed web endpoint's URL.
|
||||
|
||||
There exists an Modal LLM wrapper, which you can access with
|
||||
```python
|
||||
from langchain.llms import Modal
|
||||
```
|
||||
|
||||
endpoint_url = "https://ecorp--custom-llm-endpoint.modal.run" # REPLACE ME with your deployed Modal web endpoint's URL
|
||||
|
||||
llm = Modal(endpoint_url=endpoint_url)
|
||||
llm_chain = LLMChain(prompt=prompt, llm=llm)
|
||||
|
||||
question = "What NFL team won the Super Bowl in the year Justin Beiber was born?"
|
||||
|
||||
llm_chain.run(question)
|
||||
```
|
||||
|
||||
|
||||
@ -6,12 +6,12 @@
|
||||
"source": [
|
||||
"# Modal\n",
|
||||
"\n",
|
||||
"The [Modal Python Library](https://modal.com/docs/guide) provides convenient, on-demand access to serverless cloud compute from Python scripts on your local computer. \n",
|
||||
"The `Modal` itself does not provide any LLMs but only the infrastructure.\n",
|
||||
"The [Modal cloud platform](https://modal.com/docs/guide) provides convenient, on-demand access to serverless cloud compute from Python scripts on your local computer. \n",
|
||||
"Use `modal` to run your own custom LLM models instead of depending on LLM APIs.\n",
|
||||
"\n",
|
||||
"This example goes over how to use LangChain to interact with `Modal`.\n",
|
||||
"This example goes over how to use LangChain to interact with a `modal` HTTPS [web endpoint](https://modal.com/docs/guide/webhooks).\n",
|
||||
"\n",
|
||||
"[Here](https://modal.com/docs/guide/ex/potus_speech_qanda) is another example how to use LangChain to interact with `Modal`."
|
||||
"[_Question-answering with LangChain_](https://modal.com/docs/guide/ex/potus_speech_qanda) is another example of how to use LangChain alonside `Modal`. In that example, Modal runs the LangChain application end-to-end and uses OpenAI as its LLM API."
|
||||
]
|
||||
},
|
||||
{
|
||||
@ -22,7 +22,7 @@
|
||||
},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"!pip install modal-client"
|
||||
"!pip install modal"
|
||||
]
|
||||
},
|
||||
{
|
||||
@ -36,20 +36,14 @@
|
||||
"name": "stdout",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"\u001b[?25lLaunching login page in your browser window\u001b[33m...\u001b[0m\n",
|
||||
"\u001b[2KIf this is not showing up, please copy this URL into your web browser manually:\n",
|
||||
"\u001b[2Km⠙\u001b[0m Waiting for authentication in the web browser...\n",
|
||||
"\u001b]8;id=417802;https://modal.com/token-flow/tf-ptEuGecm7T1T5YQe42kwM1\u001b\\\u001b[4;94mhttps://modal.com/token-flow/tf-ptEuGecm7T1T5YQe42kwM1\u001b[0m\u001b]8;;\u001b\\\n",
|
||||
"\n",
|
||||
"\u001b[2K\u001b[32m⠙\u001b[0m Waiting for authentication in the web browser...\n",
|
||||
"\u001b[1A\u001b[2K^C\n",
|
||||
"\n",
|
||||
"\u001b[31mAborted.\u001b[0m\n"
|
||||
"Launching login page in your browser window...\n",
|
||||
"If this is not showing up, please copy this URL into your web browser manually:\n",
|
||||
"https://modal.com/token-flow/tf-Dzm3Y01234mqmm1234Vcu3\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"# register and get a new token\n",
|
||||
"# Register an account with Modal and get a new token.\n",
|
||||
"\n",
|
||||
"!modal token new"
|
||||
]
|
||||
@ -58,7 +52,53 @@
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Follow [these instructions](https://modal.com/docs/guide/secrets) to deal with secrets."
|
||||
"The [`langchain.llms.modal.Modal`](https://github.com/hwchase17/langchain/blame/master/langchain/llms/modal.py) integration class requires that you deploy a Modal application with a web endpoint that complies with the following JSON interface:\n",
|
||||
"\n",
|
||||
"1. The LLM prompt is accepted as a `str` value under the key `\"prompt\"`\n",
|
||||
"2. The LLM response returned as a `str` value under the key `\"prompt\"`\n",
|
||||
"\n",
|
||||
"**Example request JSON:**\n",
|
||||
"\n",
|
||||
"```json\n",
|
||||
"{\n",
|
||||
" \"prompt\": \"Identify yourself, bot!\",\n",
|
||||
" \"extra\": \"args are allowed\",\n",
|
||||
"}\n",
|
||||
"```\n",
|
||||
"\n",
|
||||
"**Example response JSON:**\n",
|
||||
"\n",
|
||||
"```json\n",
|
||||
"{\n",
|
||||
" \"prompt\": \"This is the LLM speaking\",\n",
|
||||
"}\n",
|
||||
"```\n",
|
||||
"\n",
|
||||
"An example 'dummy' Modal web endpoint function fulfilling this interface would be\n",
|
||||
"\n",
|
||||
"```python\n",
|
||||
"...\n",
|
||||
"...\n",
|
||||
"\n",
|
||||
"class Request(BaseModel):\n",
|
||||
" prompt: str\n",
|
||||
"\n",
|
||||
"@stub.function()\n",
|
||||
"@modal.web_endpoint(method=\"POST\")\n",
|
||||
"def web(request: Request):\n",
|
||||
" _ = request # ignore input\n",
|
||||
" return {\"prompt\": \"hello world\"}\n",
|
||||
"```\n",
|
||||
"\n",
|
||||
"* See Modal's [web endpoints](https://modal.com/docs/guide/webhooks#passing-arguments-to-web-endpoints) guide for the basics of setting up an endpoint that fulfils this interface.\n",
|
||||
"* See Modal's ['Run Falcon-40B with AutoGPTQ'](https://modal.com/docs/guide/ex/falcon_gptq) open-source LLM example as a starting point for your custom LLM!"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Once you have a deployed Modal web endpoint, you can pass its URL into the `langchain.llms.modal.Modal` LLM class. This class can then function as a building block in your chain."
|
||||
]
|
||||
},
|
||||
{
|
||||
@ -90,7 +130,8 @@
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"llm = Modal(endpoint_url=\"YOUR_ENDPOINT_URL\")"
|
||||
"endpoint_url = \"https://ecorp--custom-llm-endpoint.modal.run\" # REPLACE ME with your deployed Modal web endpoint's URL\n",
|
||||
"llm = Modal(endpoint_url=endpoint_url)"
|
||||
]
|
||||
},
|
||||
{
|
||||
|
||||
Loading…
Reference in New Issue
Block a user