Feat: Perplexity support

[joaoGabriel55]

Issue

#20

Description

This PR consists on add Perplexity API support

• models: https://docs.perplexity.ai/guides/model-cards;
• pricing: https://docs.perplexity.ai/guides/pricing

[joaoGabriel55]

Please, let me know if there is something missing on this first implementation

Question: How can I run tests locally?

[adenta]

How do I test?
You should be able to set API keys locally and run rspec to test the application.

@joaoGabriel55 I started a pull request here. Do you want to either tweak what I have, or incorporate my changes into your branch?

[gquaresma-godaddy]

@adenta let me know if there is something to improve

[adenta]

@gquaresma-godaddy we need tests! Want to take a crack at it?

[gquaresma-godaddy]

Sure! I will work on it.

[crmne]

Looks very good. It looks like sonar has support for vision, so it would need to be added to the chat_content_spec vision test, and the chat_streaming_spec.

[crmne]

Two new features to consider in your provider implementation:

1. Model Aliases: Please add entries for your provider in aliases.json:

   "claude-3-5-sonnet": {
     "anthropic": "claude-3-5-sonnet-20241022",
     "your-provider": "your-provider-specific-id"
   }

2. Provider Selection: Users will be able to specify your provider:

   chat = RubyLLM.chat(model: 'claude-3-5-sonnet', provider: 'your-provider')

Docs: https://rubyllm.com/guides/models#using-model-aliases

[crmne]

Added configuration requirements handling in 75f99a1

Each provider now specifies what configuration is required via a simple configuration_requirements method (you will need to implement this in your main provider file) that returns an array of config keys as symbols. The Provider module uses this to:

1. Determine if a provider is properly configured
2. Throw an error if you're trying to use that provider without configuration
3. Include ready-to-paste configuration code in the error message
4. Skip unconfigured providers during model refresh while preserving their models

Example of the new error messages:

RubyLLM::ConfigurationError: anthropic provider is not configured. Add this to your initialization:

RubyLLM.configure do |config|
  config.anthropic_api_key = ENV['ANTHROPIC_API_KEY']
end

[crmne]

@joaoGabriel55 is this still on your radar? I'd love to merge Perplexity support soon. Whenever you're ready, could you resolve the conflicts and request a review?

[crmne]

Thank you for your work!

It looks generally good, I just need some changes and tests. Please pick the cheapest model to test with.

[crmne]

hi @joaoGabriel55 are you able to provide some VCRs for this?

[joaoGabriel55]

hi @joaoGabriel55 are you able to provide some VCRs for this?
While running the tests, I got this error:

5) RubyLLM::Chat function calling perplexity/sonar can use tools with multi-turn streaming conversations
     Failure/Error: raise UnsupportedFunctionsError, "Model #{@model.id} doesn't support function calling"
     
     RubyLLM::UnsupportedFunctionsError:
       Model sonar doesn't support function calling
     # ./lib/ruby_llm/chat.rb:50:in 'RubyLLM::Chat#with_tool'
     # ./spec/ruby_llm/chat_tools_spec.rb:110:in 'block (4 levels) in <top (required)>'
     # ./spec/spec_helper.rb:86:in 'block (3 levels) in <top (required)>'
     # /Users/quaresma/.rvm/gems/ruby-3.4.1/gems/vcr-6.3.1/lib/vcr/util/variable_args_block_caller.rb:9:in 'VCR::VariableArgsBlockCaller#call_block'
     # /Users/quaresma/.rvm/gems/ruby-3.4.1/gems/vcr-6.3.1/lib/vcr.rb:194:in 'VCR#use_cassette'
     # ./spec/spec_helper.rb:85:in 'block (2 levels) in <top (required)>'
     # /Users/quaresma/.rvm/gems/ruby-3.4.1/gems/webmock-3.25.1/lib/webmock/rspec.rb:39:in 'block (2 levels) in <top (required)>'

Any ideas about that 😅 ? This is the sonar model config from the models.json file:

  {
    "id": "sonar",
    "created_at": null,
    "display_name": "Sonar",
    "provider": "perplexity",
    "context_window": 128000,
    "max_tokens": 4096,
    "type": "chat",
    "family": "sonar",
    "supports_vision": true,
    "supports_functions": false,
    "supports_json_mode": true,
    "input_price_per_million": 1.0,
    "output_price_per_million": 1.0,
    "metadata": {
      "description": "Lightweight offering with search grounding, quicker and cheaper than Sonar Pro."
    }
  },

[adenta]

@joaoGabriel55 I think if you remove perplexity from this array it shouldn't be considered a CHAT_MODEL and wont case the spec failures.

The errors are coming becasue the model doesn't support function calling and right now, the tests are assuming everything does.

[crmne]

@adenta no. It is a chat model so it stays in there. Plenty of models and providers have quirks that we already skip. Skipping means that when they fix their quirks we simply remove the skip statement. On top of that if you now have this specific model out of the CHAT_MODELS array, you need to duplicate all the tests only for Perplexity.

[crmne]

Hi @joaoGabriel55 simply skip the test in that case. There are plenty of skipped tests examples because of quirks with specific models.

[crmne]

Hey @joaoGabriel55 I feel like we're really close to the finish line here! There are only some failing tests (please make sure you have done overcommit --install before committing) and some merge conflicts.

Really looking forward to having this ship with 1.4!