instructor-clj.core — org.clojars.kapil/instructor-clj 0.0.1-alpha.2

call-llm-api clj

(call-llm-api api-url headers body)

Makes a POST request to a specified LLM API endpoint with given headers and body.

create-chat-completion clj

(create-chat-completion chat-completion-fn client-params)

(create-chat-completion chat-completion-fn client-params opts)

Creates a chat completion using OpenAI API.

This function takes OpenAI chat completion function as the first argument.

Second argument is a map with keys :messages, :model, and :response-model. :messages should be a vector of maps, each map representing a message with keys :role and :content. :model specifies the OpenAI model to use for generating completions. :response-model is a map specifying the schema and name of the response model.

Alternatively the api-key, organization, api-endpoint can be passed in the options argument of each api function. https://github.com/wkok/openai-clojure/blob/main/doc/01-usage-openai.md#options

Also, request options may be set on the underlying hato http client by adding a :request map to :options for example setting the request timeout. https://github.com/wkok/openai-clojure/blob/main/doc/01-usage-openai.md#request-options

Example: (require '[instructor-clj.core :as ic]) (require '[wkok.openai-clojure.api :as client])

(def User [:map [:name :string] [:age :int]])

(ic/create-chat-completion client {:messages [{:role "user", :content "Jason Liu is 30 years old"}] :model "gpt-3.5-turbo" :response-model User})

Returns a map with extracted information in a structured format.

default-client-params clj

instruct clj

(instruct prompt
          response-schema
          &
          {:keys [api-key _max-tokens _model _temperature max-retries]
           :as client-params
           :or {max-retries 0}})

Attempts to obtain a valid response from the LLM based on the given prompt and schema, retrying up to max-retries times if necessary.

llm->response clj

(llm->response {:keys [prompt response-schema api-key max-tokens model
                       temperature]})

The function performs the LLM call and tries to destructure and get the actual response. Returns nil in cases where the LLM is not able to generate the expected response.

@TODO Add ability to plugin different LLMs @TODO Getting response is brittle and not extensible for different LLMs

parse-generated-body clj

(parse-generated-body body)

Parses the body of a response generated by an LLM API call. Extracts and converts the message content into a Clojure map.

schema->system-prompt clj

(schema->system-prompt schema)

Converts a malli schema into JSON schema and generates a system prompt for responses