1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
|
import OpenAI from "openai";
import { MAX_TOKENS, RESPONSE_LENGTH } from "./logic/constants";
import type { AIModelAPI, ChatMessage, InputToken } from "./types";
import type { AsyncRes } from "@sortug/lib";
import type { ChatCompletionContentPart } from "openai/resources";
import { memoize } from "./cache";
import type { ChatCompletionCreateParamsNonStreaming } from "groq-sdk/src/resources/chat/completions.js";
type OChoice = OpenAI.Chat.Completions.ChatCompletion.Choice;
type Message = OpenAI.Chat.Completions.ChatCompletionUserMessageParam;
type Params = OpenAI.Chat.Completions.ChatCompletionCreateParamsNonStreaming;
type OMessage = OpenAI.Chat.Completions.ChatCompletionMessageParam;
type Props = {
baseURL: string;
apiKey: string;
model?: string;
maxTokens?: number;
tokenizer?: (text: string) => number;
allowBrowser?: boolean;
};
export default class OpenAIAPI implements AIModelAPI {
private cachedCreate!: (
args: Params,
) => Promise<OpenAI.Chat.Completions.ChatCompletion>;
private apiKey;
private baseURL;
private api;
maxTokens: number = MAX_TOKENS;
tokenizer: (text: string) => number = (text) => text.length / 3;
model;
constructor(props: Props) {
this.apiKey = props.apiKey;
this.baseURL = props.baseURL;
this.api = new OpenAI({
baseURL: this.baseURL,
apiKey: this.apiKey,
dangerouslyAllowBrowser: props.allowBrowser || false,
});
this.model = props.model || "";
if (props.maxTokens) this.maxTokens = props.maxTokens;
if (props.tokenizer) this.tokenizer = props.tokenizer;
const boundCreate = this.api.chat.completions.create.bind(
this.api.chat.completions,
);
this.cachedCreate = memoize(boundCreate, {
ttlMs: 2 * 60 * 60 * 1000, // 2h
maxEntries: 5000,
persistDir: "./cache/memo",
// stable key for the call
keyFn: (args) => {
// args is the single object param to .create(...)
const {
model,
messages,
max_tokens,
temperature,
top_p,
frequency_penalty,
presence_penalty,
stop,
} = args as Params;
// stringify messages deterministically (role+content only)
const msg = (messages as any[])
.map((m) => ({ role: m.role, content: m.content }))
.slice(0, 200); // guard size if you want
return JSON.stringify({
model,
msg,
max_tokens,
temperature,
top_p,
frequency_penalty,
presence_penalty,
stop,
});
},
});
}
public setModel(model: string) {
this.model = model;
}
private mapMessages(input: ChatMessage[]): Message[] {
return input.map((m) => {
return { role: m.author as any, content: m.text, name: m.author };
});
}
private buildInput(tokens: InputToken[]): Message[] {
const content: ChatCompletionContentPart[] = tokens.map((t) => {
if ("text" in t) return { type: "text", text: t.text };
if ("img" in t) return { type: "image_url", image_url: { url: t.img } };
else return { type: "text", text: "oy vey" };
});
return [{ role: "user", content }];
}
public async send(
input: string | InputToken[],
sys?: string,
): AsyncRes<string> {
const messages: Message[] =
typeof input === "string"
? [{ role: "user" as const, content: input }]
: this.buildInput(input);
// const messages = this.mapMessages(input);
const allMessages: OMessage[] = sys
? [{ role: "system", content: sys }, ...messages]
: messages;
const truncated = this.truncateHistory(allMessages);
const res = await this.apiCall(truncated);
if ("error" in res) return res;
else {
try {
// TODO type this properly
const choices: OChoice[] = res.ok;
const resText = choices.reduce((acc, item) => {
return `${acc}\n${item.message.content || ""}`;
}, "");
return { ok: resText };
} catch (e) {
return { error: `${e}` };
}
}
}
public async stream(
input: string | InputToken[],
handle: (c: string) => void,
sys?: string,
) {
const messages: Message[] =
typeof input === "string"
? [{ role: "user" as const, content: input }]
: this.buildInput(input);
// const messages = this.mapMessages(input);
const allMessages: OMessage[] = sys
? [{ role: "system", content: sys }, ...messages]
: messages;
const truncated = this.truncateHistory(allMessages);
await this.apiCallStream(truncated, handle);
}
private truncateHistory(messages: OMessage[]): OMessage[] {
const totalTokens = messages.reduce((total, message) => {
return total + this.tokenizer(message.content as string);
}, 0);
while (totalTokens > this.maxTokens && messages.length > 1) {
// Always keep the system message if it exists
const startIndex = messages[0].role === "system" ? 1 : 0;
messages.splice(startIndex, 1);
}
return messages;
}
// TODO custom temperature?
private async apiCall(messages: OMessage[]): AsyncRes<OChoice[]> {
// console.log({ messages }, "at the very end");
try {
const completion = await this.cachedCreate({
// temperature: 1.3,
model: this.model,
messages,
max_tokens: RESPONSE_LENGTH,
});
if (!completion) return { error: "null response from openai" };
return { ok: completion.choices };
} catch (e) {
console.log(e, "error in openai api");
return { error: `${e}` };
}
}
private async apiCallStream(
messages: OMessage[],
handle: (c: string) => void,
): Promise<void> {
try {
const stream = await this.api.chat.completions.create({
temperature: 1.3,
model: this.model,
messages,
max_tokens: RESPONSE_LENGTH,
stream: true,
});
for await (const chunk of stream) {
for (const choice of chunk.choices) {
console.log({ choice });
if (!choice.delta) continue;
const cont = choice.delta.content;
if (!cont) continue;
handle(cont);
}
}
} catch (e) {
console.log(e, "error in openai api");
handle(`Error streaming OpenAI, ${e}`);
}
}
}
|
