Anyscale

Integrate Anyscale endpoints with Portkey seamlessly and make your OSS models production-ready

Portkey's suite of features - AI gateway, observability, prompt management, and continuous fine-tuning are all enabled for the OSS models (Llama2, Mistral, Zephyr, and more) available on Anyscale endpoints.

Portkey SDK Integration with Anyscale

1. Install the Portkey SDK

NodeJS

npm install --save portkey-ai

Python

pip install portkey-ai

2. Initialize Portkey with Anyscale Virtual Key

Set up the Portkey instance with your Anyscale virtual key as part of the initialization configuration. (Create virtual key for Anyscale here)

NodeJS SDK

import Portkey from 'portkey-ai'
 
const portkey = new Portkey({
    apiKey: "PORTKEY_API_KEY", // defaults to process.env["PORTKEY_API_KEY"]
    virtualKey: "ANYSCALE_VIRTUAL_KEY" // Your Anyscale Virtual Key
})

Python SDK

from portkey_ai import Portkey

portkey = Portkey(
    api_key="PORTKEY_API_KEY",  # Replace with your Portkey API key
    virtual_key="ANYSCALE_VIRTUAL_KEY"   # Replace with your virtual key for Anyscale
)

3. Invoke Chat Completions with Anyscale

NodeJS SDK

const chatCompletion = await portkey.chat.completions.create({
    messages: [{ role: 'user', content: 'Say this is a test' }],
    model: 'mistralai/Mistral-7B-Instruct-v0.1',
});

console.log(chatCompletion.choices);

Python SDK

completion = portkey.chat.completions.create(
    messages= [{ "role": 'user', "content": 'Say this is a test' }],
    model= 'mistralai/Mistral-7B-Instruct-v0.1'
)

print(completion.choices)

Directly Using Portkey's REST API

Alternatively, you can also directly call Anyscale models through Portkey's REST API - it works exactly the same as OpenAI API, with 2 differences:

1. You send your requests to Portkey's complete Gateway URL https://api.portkey.ai/v1/chat/completions

2. You have to add Portkey specific headers.

   1. x-portkey-api-key for sending your Portkey API Key

   2. x-portkey-virtual-key for sending your provider's virtual key (Alternatively, if you are not using Virtual keys, you can send your Auth header for your provider, and pass the x-portkey-provider header along with it)

REST API

curl https://api.portkey.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $ANYSCALE_API_KEY" \
  -H "x-portkey-api-key: $PORTKEY_API_KEY" \
  -H "x-portkey-provider: anyscale" \ 
  -d '{
    "model": "mistralai/Mistral-7B-Instruct-v0.1",
    "messages": [{"role": "user","content": "Hello!"}]
  }'

List of all possible Portkey headers.

Using the OpenAI Python or Node SDKs for Anyscale

You can also use the baseURL param in the standard OpenAI SDKs and make calls to Portkey + Anyscale directly from there. Like the Rest API example, you are only required to change the baseURL and add defaultHeaders to your instance. You can use the Portkey SDK to make it simpler:

OpenAI Node SDK

import OpenAI from 'openai'; // We're using the v4 SDK
import { PORTKEY_GATEWAY_URL, createHeaders } from 'portkey-ai'

const anyscale = new OpenAI({
  apiKey: 'ANYSCALE_API_KEY',
  baseURL: PORTKEY_GATEWAY_URL,
  defaultHeaders: createHeaders({
    provider: "anyscale",
    apiKey: "PORTKEY_API_KEY" // defaults to process.env["PORTKEY_API_KEY"]
  })
});

async function main() {
  const chatCompletion = await anyscale.chat.completions.create({
    messages: [{ role: 'user', content: 'Say this is a test' }],
    model: 'mistralai/Mistral-7B-Instruct-v0.1',
  });

  console.log(chatCompletion.choices);
}

main();

OpenAI Python SDK

from openai import OpenAI
from portkey_ai import PORTKEY_GATEWAY_URL, createHeaders

anyscale = OpenAI(
    api_key="ANYSCALE_API_KEY", # defaults to os.environ.get("OPENAI_API_KEY")
    base_url=PORTKEY_GATEWAY_URL,
    default_headers=createHeaders(
        provider="anyscale",
        api_key="PORTKEY_API_KEY" # defaults to os.environ.get("PORTKEY_API_KEY")
    )
)

chat_complete = anyscale.chat.completions.create(
    model="mistralai/Mistral-7B-Instruct-v0.1",
    messages=[{"role": "user", "content": "Say this is a test"}],
)

print(chat_complete.choices[0].message.content)

This request will be automatically logged by Portkey. You can view this in your logs dashboard. Portkey logs the tokens utilized, execution time, and cost for each request. Additionally, you can delve into the details to review the precise request and response data.

Managing Anyscale Prompts

You can manage all prompts for Anyscale's OSS models in the Prompt Library. All the current models of Anyscale are supported.

Creating Prompts

Use the Portkey prompt playground to set variables and try out various model params to get the right output.

Using Prompts

Deploy the prompts using the Portkey SDK or REST API

Node SDK

import Portkey from 'portkey-ai'

const portkey = new Portkey({
    apiKey: "PORTKEY_API_KEY", // defaults to process.env["PORTKEY_API_KEY"]
})

// Make the prompt creation call with the variables
const promptCompletion = await portkey.prompts.completions.create({
    promptID: "YOUR_PROMPT_ID",
    variables: {
        //Required variables for prompt
    }
})

We can also override the hyperparameters:

const promptCompletion = await portkey.prompts.completions.create({
    promptID: "YOUR_PROMPT_ID",
    variables: {
        //Required variables for prompt
    },
    max_tokens: 250,
    presence_penalty: 0.2
})

Python SDK

from portkey_ai import Portkey

client = Portkey(
    api_key="PORTKEY_API_KEY",  # defaults to os.environ.get("PORTKEY_API_KEY")
)

prompt_completion = client.prompts.completions.create(
    prompt_id="YOUR_PROMPT_ID",
    variables={
        #Required variables for prompt
    }
)

print(prompt_completion.data)

We can also override the hyperparameters:

prompt_completion = client.prompts.completions.create(
    prompt_id="YOUR_PROMPT_ID",
    variables={
        #Required variables for prompt
    },
    max_tokens=250,
    presence_penalty=0.2
)
print(prompt_completion.data)

REST API

curl -X POST "https://api.portkey.ai/v1/prompts/9218b4e6-52db-41a4-b963-4ee6505ed758/completions" \
-H "Content-Type: application/json" \
-H "x-portkey-api-key: $PORTKEY_API_KEY" \
-d '{
    "variables": {
        "title": "The impact of AI on middle school teachers",
        "num_sections": "5"
    },
    "max_tokens": 250, # Optional
    "presence_penalty": 0.2 # Optional
}'

Observe how this streamlines your code readability and simplifies prompt updates via the UI without altering the codebase.

---

List of Models Supported

Model Name	Model Key on Portkey
Llama 2 - 7B	meta-llama/Llama-2-7b-chat-hf
Llama 2- 13B	meta-llama/Llama-2-13b-chat-hf
Llama 2 - 70B	meta-llama/Llama-2-70b-chat-hf
Code Llama - 34B	codellama/CodeLlama-34b-Instruct-hf
Mistral - 7B	mistralai/Mistral-7B-Instruct-v0.1
HF Zephyr - 7B (Based on Mistral-7B)	HuggingFaceH4/zephyr-7b-beta

---

Advanced Use Cases

Streaming Responses

Portkey supports streaming responses using Server Sent Events (SSE).

OpenAI NodeJS SDK

import OpenAI from 'openai';
import { PORTKEY_GATEWAY_URL, createHeaders } from 'portkey-ai'

const anyscale = new OpenAI({
  baseURL: PORTKEY_GATEWAY_URL,
  defaultHeaders: createHeaders({
    mode: "anyscale",
    apiKey: "PORTKEY_API_KEY" // defaults to process.env["PORTKEY_API_KEY"]
  })
});

async function main() {
  const stream = await anyscale.chat.completions.create({
    model: 'mistralai/Mistral-7B-Instruct-v0.1',
    messages: [{ role: 'user', content: 'Say this is a test' }],
    stream: true,
  });
  for await (const chunk of stream) {
    process.stdout.write(chunk.choices[0]?.delta?.content || '');
  }
}

main();

OpenAI Python SDK

from openai import OpenAI
from portkey_ai import PORTKEY_GATEWAY_URL, createHeaders

anyscale = OpenAI(
    api_key="ANYSCALE-API-KEY",  # defaults to os.environ.get("OPENAI_API_KEY")
    base_url=PORTKEY_GATEWAY_URL,
    default_headers=createHeaders(
        provider="anyscale",
        api_key="PORTKEY-API-KEY" # defaults to os.environ.get("PORTKEY_API_KEY")
    )
)

chat_complete = anyscale.chat.completions.create(
    model="mistralai/Mistral-7B-Instruct-v0.1",
    messages=[{"role": "user", "content": "Say this is a test"}],
    stream=True
)

for chunk in chat_complete:
    print(chunk.choices[0].delta.content, end="", flush=True)

Fine-tuning

Please refer to our fine-tuning guides to take advantage of Portkey's advanced continuous fine-tuning capabilities.

Portkey Features

Portkey supports the complete host of it's functionality via the OpenAI SDK so you don't need to migrate away from it.

Please find more information in the relevant sections:

1. Add metadata to your requests

2. Add gateway configs to the client or a single request

3. Trace Anyscale requests

4. Setup a fallback to Azure OpenAI