This is a reverse engineered API wrapper for the Vercel AI Playground, which allows for free access to many LLMs, including OpenAI's ChatGPT, Cohere's Command Nightly, as well as some open source models.
Also a JavaScript implementation of the ading2210's vercel-llm-api Python library.
Table of contents generated with markdown-toc
- Download the available models
- Generate text
- Generate chat messages
- Set custom parameters
- Stream the responses
- No auth support
- Can't use "pro" or "hobby" models
You can install this library by running the following command:
npm install vercel-llm-api
Note that the entire library requires the use of async/await.
To use this library, simply require('vercel-llm-api')
and create a Client
instance. You can specify custom Axios request configurations as an argument.
See here for the Axios request config.
const { Client } = require('vercel-llm-api'),
client = new Client();
client.on('ready', async () => {
// the client is ready to do whatever
});
Note that the following examples assume client
is the name of your Client
instance and that it is inside an async function.
The client downloads the available models upon initialization, and stores them in client.models
.
>>> console.log(client.models)
{
"anthropic:claude-instant-v1": {
"id": "anthropic:claude-instant-v1", // the model's id
"provider": "anthropic", // the model's provider
"providerHumanName": "Anthropic", // the provider's display name
"makerHumanName": "Anthropic", // the maker of the model
"minBillingTier": "hobby", // the minimum billing tier needed to use the model
"parameters": { // an object of optional parameters that can be passed to the generate function
"temperature": { // the name of the parameter
"value": 1, // the default value for the parameter
"range": [0, 1] // a range of possible values for the parameter
},
...
}
...
}
}
Note that, since there is no auth yet, if a model has the "minBillingTier"
property present, it can't be used.
A list of model IDs is also available in client.model_ids
.
>>> console.log(client.model_ids)
[
"anthropic:claude-instant-v1", // locked to hobby tier; unusable
"anthropic:claude-v1", // locked to hobby tier; unusable
"replicate:replicate/alpaca-7b",
"replicate:stability-ai/stablelm-tuned-alpha-7b",
"huggingface:bigscience/bloom",
"huggingface:bigscience/bloomz",
"huggingface:google/flan-t5-xxl",
"huggingface:google/flan-ul2",
"huggingface:EleutherAI/gpt-neox-20b",
"huggingface:OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5",
"huggingface:bigcode/santacoder",
"cohere:command-medium-nightly",
"cohere:command-xlarge-nightly",
"openai:gpt-4", // locked to pro tier; unusable
"openai:code-cushman-001",
"openai:code-davinci-002",
"openai:gpt-3.5-turbo",
"openai:text-ada-001",
"openai:text-babbage-001",
"openai:text-curie-001",
"openai:text-davinci-002",
"openai:text-davinci-003"
]
An Object of default parameters for each model can be found at client.model_params
.
>>> console.log(client.model_defaults)
{
"anthropic:claude-instant-v1": {
"temperature": 1,
"maximumLength": 200,
"topP": 1,
"topK": 1,
"presencePenalty": 1,
"frequencyPenalty": 1,
"stopSequences": [
"\n\nHuman:"
]
},
...
}
To generate some text, use the client.generate
function, which accepts the following arguments:
model
- The ID of the model you want to use.prompt
- Your prompt.params
- An Object of optional parameters. See the previous section for how to find these.
The function returns the newly generated text as a ReadableStream
.
await client.generate('openai:gpt-3.5-turbo', 'Summarize the GNU GPL v3');
To generate chat messages, use the client.chat
function, which accepts the following arguments:
model
- The ID of the model you want to use.messages
- A list of messages. The format for this is identical to how you would use the official OpenAI API.params
- An Object of optional parameters. See the "Downloading the Available Models" section for how to find these.
The function returns the newly generated text as a ReadableStream
.
const messages = [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "Who won the world series in 2020?"},
{"role": "assistant", "content": "The Los Angeles Dodgers won the World Series in 2020."},
{"role": "user", "content": "Where was it played?"}
];
await client.chat('openai:gpt-3.5-turbo', messages);
StreamHandler is a utility function to handle the returned ReadableStream
of the instantiated Client
's chat
and generate
functions.
StreamHandler accepts the following arguments:
stream
- TheReadableStream
.callback
- An optional callback to process each chunk of the stream.
...and returns an Array
of String
s.
const { Client, StreamHandler } = require('vercel-llm-api'),
client = new Client();
client.on('ready', async () => {
const messages = [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "Who won the world series in 2020?"},
{"role": "assistant", "content": "The Los Angeles Dodgers won the World Series in 2020."},
{"role": "user", "content": "Where was it played?"}
];
const stream = await client.chat('openai:gpt-3.5-turbo', messages),
response = await StreamHandler(stream);
console.log(response); // returns [ "The", " 2020", " World", " Series", " was", " played", ... ]
});
If you want to show the debug messages, simply listen to the debug
event of the Client
instance.
const { Client } = require('vercel-llm-api'),
client = new Client();
client.on('debug', console.log);