openai: add stuff

.changeset/calm-horses-impress.md

@@ -0,0 +1,5 @@
+---
+'@paperdave/openai': minor
+---
+
+Add Text Edit

.changeset/late-pumas-visit.md

@@ -0,0 +1,5 @@
+---
+'@paperdave/openai': minor
+---
+
+Implement Retry Count

.changeset/slow-fishes-chew.md

@@ -0,0 +1,5 @@
+---
+'@paperdave/openai': minor
+---
+
+Move to using camel case everywhere

packages/openai/README.md

@@ -82,16 +82,15 @@ const response = await generateChatCompletion({
 
  // Optional: See OpenAI Docs
  stream: boolean,
- max_tokens: number,
+ maxTokens: number,
  temperature: number,
- top_p: number,
+ topP: number,
  n: number,
- logprobs: number,
  stop: string | string[],
- presence_penalty: number,
- frequency_penalty: number,
- best_of: number,
- logit_bias: object,
+ presencePenalty: number,
+ frequencyPenalty: number,
+ bestOf: number,
+ logitBias: Record<string, number>,
  user: string,
 });
 ```
@@ -171,13 +170,75 @@ interface ChatCompletionMultiStream {
 }
 ```
 
-## Text Completions
+## Text Completions / Insertions
 
-Implemented, Docs coming soon.
+Text completions allow you to complete a text prompt. [OpenAI's Guide](https://platform.openai.com/docs/guides/completion)
+
+One function is given to call this api: `generateTextCompletion`, which functions almost identically to `generateChatCompletion`, except that instead of a `messages` array, you pass a `prompt` string.
+
+```ts
+import { generateTextCompletion } from '@paperdave/openai';
+
+const response = await generateTextCompletion({
+ // Required arguments, see OpenAI Docs
+ model: TextModel,
+ prompt: string,
+
+ // Optionally override API key and organization.
+ auth: AuthOverride,
+ // Number of times to retry due to network/ratelimit issues, default 3
+ retry: number,
+
+ // Optional: See OpenAI Docs
+ suffix: string,
+ max_tokens: number,
+ temperature: number,
+ topP: number,
+ n: number,
+ logProbs: number,
+ stop: string | string[],
+ echo: boolean,
+ presencePenalty: number,
+ frequencyPenalty: number,
+ bestOf: number,
+ logitBias: Record<string, number>,
+ user: string,
+});
+```
+
+The return type is almost identical to `generateChatCompletion`. To understand exactly how it works and how to stream results, see the [Chat Completions](#chat-completions) section above.
+
+In addition, you can pass `logProbs: number` to get a `logProbs` object on the response.
 
 ## Text Edits
 
-Coming Soon
+The edits endpoint can be used to edit text, rather than just completing it. [OpenAI's Guide](https://platform.openai.com/docs/guides/completion/editing-text)
+
+Editing text is done with the `generateTextEdit` function, which accepts the following arguments:
+
+```ts
+import { generateTextEdit } from '@paperdave/openai';
+
+const response = await generateTextEdit({
+ // Required arguments, see OpenAI Docs
+ model: TextEditModel,
+ input?: string,
+ instruction: string,
+
+ // Optionally override API key and organization.
+ auth: AuthOverride,
+ // Number of times to retry due to network/ratelimit issues, default 3
+ retry: number,
+
+ // Optional: See OpenAI Docs
+ temperature: number,
+ topP: number,
+ n: number,
+ // user: string, // OpenAI should have this, but they dont
+});
+```
+
+Other than the lack of streaming, the return type is nearly identical to `generateChatCompletion`. To understand exactly how it works, see the [Chat Completions](#chat-completions) section above.
 
 ## Images
 

packages/openai/demos/edit-text.ts

@@ -0,0 +1,8 @@
+import { generateTextEdit } from '../src/text-edit';
+
+const generation = await generateTextEdit({
+ model: 'text-davinci-edit-001',
+ input: 'The quick brown fox jumps over the lazy dog.',
+ instruction: 'Convert to leet speak.',
+});
+console.log(generation);

packages/openai/demos/text-stream.ts → packages/openai/demos/text-gen-stream.ts

@@ -2,10 +2,10 @@ import { generateTextCompletion } from '../src/text';
 
 const stream = await generateTextCompletion({
  model: 'ada',
- prompt: 'say this is a test.',
+ prompt: '',
  stream: true,
- max_tokens: 10,
- temperature: 1.25,
+ max_tokens: 500,
+ temperature: 2,
 });
 
 for await (const text of stream.content) {

packages/openai/src/chat.ts

@@ -1,7 +1,8 @@
 import { createArray, IterableStream } from '@paperdave/utils';
-import { AuthOverride, getAuthHeaders } from './api-key';
+import { AuthOverride } from './api-key';
+import { fetchOpenAI, FetchOptions } from './fetch';
 import { ChatModel, PRICING_CHAT, PRICING_TEXT } from './models';
-import { FinishReason, RawCompletionUsage, td } from './shared';
+import { CompletionUsage, FinishReason, RawCompletionUsage, td } from './shared';
 import { countChatPromptTokens, countTokens } from './tokenization';
 
 export type GPTMessageRole = 'system' | 'user' | 'assistant';
@@ -46,10 +47,8 @@ export interface RawChatCompletionChunkChoice {
  index: number;
 }
 
-export interface ChatCompletionOptions<
- Stream extends boolean = boolean,
- N extends number = number
-> {
+export interface ChatCompletionOptions<Stream extends boolean = boolean, N extends number = number>
+ extends FetchOptions {
  /**
  * ID of the model to use. You can use the [List models](/docs/api-reference/models/list) API to
  * see all of your available models, or see our [Model overview](/docs/models/overview) for
@@ -72,7 +71,7 @@ export interface ChatCompletionOptions<
  * your prompt plus `max_tokens` cannot exceed the model's context length. Most models have a
  * context length of 2048 tokens (except for the newest models, which support 4096).
  */
- max_tokens?: number | null;
+ maxTokens?: number | null;
  /**
  * What sampling temperature to use, between 0 and 2. Higher values like 0.8 will make the output
  * more random, while lower values like 0.2 will make it more focused and deterministic. We
@@ -85,7 +84,7 @@ export interface ChatCompletionOptions<
  * the top 10% probability mass are considered. We generally recommend altering this or
  * `temperature` but not both.
  */
- top_p?: number | null;
+ topP?: number | null;
  /**
  * How many completions to generate for each prompt. **Note:** Because this parameter generates
  * many completions, it can quickly consume your token quota. Use carefully and ensure that you
@@ -99,14 +98,14 @@ export interface ChatCompletionOptions<
  * in the text so far, increasing the model's likelihood to talk about new topics. [See more
  * information about frequency and presence penalties.](/docs/api-reference/parameter-details)
  */
- presence_penalty?: number | null;
+ presencePenalty?: number | null;
  /**
  * Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing
  * frequency in the text so far, decreasing the model's likelihood to repeat the same line
  * verbatim. [See more information about frequency and presence
  * penalties.](/docs/api-reference/parameter-details)
  */
- frequency_penalty?: number | null;
+ frequencyPenalty?: number | null;
  /**
  * Modify the likelihood of specified tokens appearing in the completion. Accepts a json object
  * that maps tokens (specified by their token ID in the GPT tokenizer) to an associated bias value
@@ -117,7 +116,7 @@ export interface ChatCompletionOptions<
  * should result in a ban or exclusive selection of the relevant token. As an example, you can
  * pass `{\"50256\": -100}` to prevent the <|endoftext|> token from being generated.
  */
- logit_bias?: Record<string, number> | null;
+ logitBias?: Record<string, number> | null;
  /**
  * A unique identifier representing your end-user, which can help OpenAI to monitor and detect
  * abuse. [Learn more](/docs/guides/safety-best-practices/end-user-ids).
@@ -134,13 +133,6 @@ export interface ChatCompletionMetadata {
  usage: CompletionUsage;
 }
 
-export interface CompletionUsage {
- promptTokens: number;
- completionTokens: number;
- totalTokens: number;
- price: number;
-}
-
 export interface ChatCompletion extends ChatCompletionMetadata, ChatCompletionChoice {}
 
 export interface ChatCompletionMulti extends ChatCompletionMetadata {
@@ -176,22 +168,28 @@ export type ChatCompletionResultFromOptions<
 export async function generateChatCompletion<Stream extends boolean, N extends number>(
  options: ChatCompletionOptions<Stream, N>
 ): Promise<ChatCompletionResultFromOptions<Stream, N>> {
- const { retry: retryCount, auth, ...gptOptions } = options;
+ const { retry, auth, ...gptOptions } = options;
 
- const response = await fetch('https://api.openai.com/v1/chat/completions', {
- method: 'post',
- headers: {
- 'Content-Type': 'application/json',
- ...getAuthHeaders(auth),
+ const response = await fetchOpenAI({
+ endpoint: '/chat/completions',
+ method: 'POST',
+ body: {
+ model: gptOptions.model,
+ prompt: gptOptions.messages,
+ max_tokens: gptOptions.maxTokens ?? (gptOptions as any).max_tokens,
+ temperature: gptOptions.temperature,
+ top_p: gptOptions.topP ?? (gptOptions as any).top_p,
+ n: gptOptions.n,
+ stop: gptOptions.stop,
+ presence_penalty: gptOptions.presencePenalty ?? (gptOptions as any).presence_penalty,
+ frequency_penalty: gptOptions.frequencyPenalty ?? (gptOptions as any).frequency_penalty,
+ logit_bias: gptOptions.logitBias ?? (gptOptions as any).logit_bias,
+ user: gptOptions.user,
  },
- body: JSON.stringify(gptOptions),
+ auth,
+ retry,
  });
 
- if (!response.ok) {
- const error = await response.json();
- throw new Error(`${response.status} ${response.statusText}: ${JSON.stringify(error)}`);
- }
-
  if (gptOptions.stream) {
  const reader = (response as any).body.getReader();
  if (!reader) {

packages/openai/src/fetch.ts

@@ -0,0 +1,64 @@
+import { delay } from '@paperdave/utils';
+import { AuthOverride, getAuthHeaders } from './api-key';
+
+export interface FetchOptions {
+ /** Number of retries before giving up. Defaults to 3. */
+ retry?: number;
+ /** Override authentication settings. */
+ auth?: AuthOverride;
+}
+
+export interface InternalFetchOptions extends FetchOptions {
+ endpoint: string;
+ method: 'GET' | 'POST' | 'PUT' | 'DELETE';
+ body?: any;
+ init?: RequestInit;
+}
+
+const base = 'https://api.openai.com/v1';
+
+export async function fetchOpenAI(options: InternalFetchOptions) {
+ const { endpoint, method, body, init, retry = 3, auth } = options;
+ const url = `${base}${endpoint}`;
+ const headers = {
+ ...(body ? { 'Content-Type': 'application/json' } : {}),
+ ...getAuthHeaders(auth),
+ };
+ const fullInit = {
+ ...init,
+ method,
+ headers: {
+ ...headers,
+ ...init?.headers,
+ },
+ body: body ? JSON.stringify(body) : undefined,
+ };
+ let tries = 0;
+ while (true) {
+ try {
+ const response = await fetch(url, fullInit);
+ if (response.status === 429) {
+ const retryAfter = response.headers.get('Retry-After');
+ if (retryAfter) {
+ await delay(parseInt(retryAfter, 10) * 1000);
+ } else {
+ await delay(1000);
+ }
+ continue;
+ }
+ if (response.status >= 400) {
+ const error = await response.json();
+ throw new Error(
+ (error as any)?.error.message ??
+ (typeof error === 'string' ? error : JSON.stringify(error))
+ );
+ }
+ return response;
+ } catch (error) {
+ if (tries >= retry) {
+ throw error;
+ }
+ tries++;
+ }
+ }
+}

packages/openai/src/index.ts

@@ -1,6 +1,8 @@
 export * from './api-key';
+export type { FetchOptions } from './fetch';
 export * from './chat';
 export * from './log-probs';
 export * from './models';
 export * from './text';
+export * from './text-edit';
 export * from './tokenization';

packages/openai/src/log-probs.ts

@@ -21,8 +21,6 @@ export class LogProbs {
  this.topLogProbs = raw.top_logprobs;
  this.textOffsets = raw.text_offset;
  }
-
- // TODO: methods to make this more interesting
 }
 
 // hidden please

packages/openai/src/models.ts

@@ -20,9 +20,16 @@ export const PRICING_TEXT = {
  babbage: 0.0005 / 1000,
  ada: 0.0004 / 1000,
 };
-
 export type TextModel = keyof typeof PRICING_TEXT;
 
+/** "During this initial beta period, usage of the edits endpoint is free." */
+export const PRICING_TEXT_EDIT = {
+ 'text-davinci-edit-001': 0,
+ 'code-davinci-edit-001': 0,
+};
+
+export type TextEditModel = keyof typeof PRICING_TEXT_EDIT;
+
 export const PRICING_IMAGE = {
  1024: 0.02,
  512: 0.018,

packages/openai/src/shared.ts

@@ -1,9 +1,9 @@
 export type FinishReason = 'stop' | 'length';
 
 export interface RawCompletionUsage {
- prompt_tokens: 10;
- completion_tokens: 10;
- total_tokens: 20;
+ prompt_tokens: number;
+ completion_tokens: number;
+ total_tokens: number;
 }
 
 export interface CompletionUsage {