forked from intitni/CopilotForXcode
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathChatGPTService.swift
More file actions
554 lines (492 loc) · 19.4 KB
/
ChatGPTService.swift
File metadata and controls
554 lines (492 loc) · 19.4 KB
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
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
import AsyncAlgorithms
import Foundation
import Preferences
public protocol ChatGPTServiceType {
var memory: ChatGPTMemory { get set }
var configuration: ChatGPTConfiguration { get set }
func send(content: String, summary: String?) async throws -> AsyncThrowingStream<String, Error>
func stopReceivingMessage() async
}
public enum ChatGPTServiceError: Error, LocalizedError {
case chatModelNotAvailable
case embeddingModelNotAvailable
case endpointIncorrect
case responseInvalid
case otherError(String)
public var errorDescription: String? {
switch self {
case .chatModelNotAvailable:
return "Chat model is not available, please add a model in the settings."
case .embeddingModelNotAvailable:
return "Embedding model is not available, please add a model in the settings."
case .endpointIncorrect:
return "ChatGPT endpoint is incorrect"
case .responseInvalid:
return "Response is invalid"
case let .otherError(content):
return content
}
}
}
public struct ChatGPTError: Error, Codable, LocalizedError {
public var error: ErrorContent
public init(error: ErrorContent) {
self.error = error
}
public struct ErrorContent: Codable {
public var message: String
public var type: String
public var param: String?
public var code: String?
public init(message: String, type: String, param: String? = nil, code: String? = nil) {
self.message = message
self.type = type
self.param = param
self.code = code
}
}
public var errorDescription: String? {
error.message
}
}
public class ChatGPTService: ChatGPTServiceType {
public var memory: ChatGPTMemory
public var configuration: ChatGPTConfiguration
public var functionProvider: ChatGPTFunctionProvider
var uuidGenerator: () -> String = { UUID().uuidString }
var cancelTask: Cancellable?
var buildCompletionStreamAPI: CompletionStreamAPIBuilder = OpenAICompletionStreamAPI.init
var buildCompletionAPI: CompletionAPIBuilder = OpenAICompletionAPI.init
public init(
memory: ChatGPTMemory = AutoManagedChatGPTMemory(
systemPrompt: "",
configuration: UserPreferenceChatGPTConfiguration(),
functionProvider: NoChatGPTFunctionProvider()
),
configuration: ChatGPTConfiguration = UserPreferenceChatGPTConfiguration(),
functionProvider: ChatGPTFunctionProvider = NoChatGPTFunctionProvider()
) {
self.memory = memory
self.configuration = configuration
self.functionProvider = functionProvider
}
/// Send a message and stream the reply.
public func send(
content: String,
summary: String? = nil
) async throws -> AsyncThrowingStream<String, Error> {
if !content.isEmpty || summary != nil {
let newMessage = ChatMessage(
id: uuidGenerator(),
role: .user,
content: content,
name: nil,
functionCall: nil,
summary: summary
)
await memory.appendMessage(newMessage)
}
return Debugger.$id.withValue(.init()) {
AsyncThrowingStream<String, Error> { continuation in
let task = Task(priority: .userInitiated) {
do {
var functionCall: ChatMessage.FunctionCall?
var functionCallMessageID = ""
var isInitialCall = true
loop: while functionCall != nil || isInitialCall {
try Task.checkCancellation()
isInitialCall = false
if let call = functionCall {
if !configuration.runFunctionsAutomatically {
break loop
}
functionCall = nil
await runFunctionCall(call, messageId: functionCallMessageID)
}
let stream = try await sendMemory()
#if DEBUG
var reply = ""
#endif
for try await content in stream {
try Task.checkCancellation()
switch content {
case let .text(text):
continuation.yield(text)
#if DEBUG
reply.append(text)
#endif
case let .functionCall(call):
if functionCall == nil {
functionCallMessageID = uuidGenerator()
functionCall = call
} else {
functionCall?.name.append(call.name)
functionCall?.arguments.append(call.arguments)
}
await prepareFunctionCall(
call,
messageId: functionCallMessageID
)
}
}
#if DEBUG
Debugger.didReceiveResponse(content: reply)
#endif
}
#if DEBUG
Debugger.didFinish()
#endif
continuation.finish()
} catch {
continuation.finish(throwing: error)
}
}
continuation.onTermination = { _ in
task.cancel()
}
}
}
}
/// Send a message and get the reply in return.
public func sendAndWait(
content: String,
summary: String? = nil
) async throws -> String? {
if !content.isEmpty || summary != nil {
let newMessage = ChatMessage(
id: uuidGenerator(),
role: .user,
content: content,
summary: summary
)
await memory.appendMessage(newMessage)
}
return try await Debugger.$id.withValue(.init()) {
let message = try await sendMemoryAndWait()
var finalResult = message?.content
var functionCall = message?.functionCall
while let call = functionCall {
try Task.checkCancellation()
if !configuration.runFunctionsAutomatically {
break
}
functionCall = nil
await runFunctionCall(call)
guard let nextMessage = try await sendMemoryAndWait() else { break }
finalResult = nextMessage.content
functionCall = nextMessage.functionCall
}
#if DEBUG
Debugger.didReceiveResponse(content: finalResult ?? "N/A")
Debugger.didFinish()
#endif
return finalResult
}
}
public func stopReceivingMessage() {
cancelTask?()
cancelTask = nil
}
}
// - MARK: Internal
extension ChatGPTService {
enum StreamContent {
case text(String)
case functionCall(ChatMessage.FunctionCall)
}
/// Send the memory as prompt to ChatGPT, with stream enabled.
func sendMemory() async throws -> AsyncThrowingStream<StreamContent, Error> {
await memory.refresh()
guard let model = configuration.model else {
throw ChatGPTServiceError.chatModelNotAvailable
}
guard let url = URL(string: configuration.endpoint) else {
throw ChatGPTServiceError.endpointIncorrect
}
let messages = await memory.messages.map {
CompletionRequestBody.Message(
role: $0.role,
content: $0.content ?? "",
name: $0.name,
function_call: $0.functionCall.map {
.init(name: $0.name, arguments: $0.arguments)
}
)
}
let remainingTokens = await memory.remainingTokens
let requestBody = CompletionRequestBody(
model: model.info.modelName,
messages: messages,
temperature: configuration.temperature,
stream: true,
stop: configuration.stop.isEmpty ? nil : configuration.stop,
max_tokens: maxTokenForReply(
maxToken: model.info.maxTokens,
remainingTokens: remainingTokens
),
function_call: model.info.supportsFunctionCalling
? functionProvider.functionCallStrategy
: nil,
functions:
model.info.supportsFunctionCalling
? functionProvider.functions.map {
ChatGPTFunctionSchema(
name: $0.name,
description: $0.description,
parameters: $0.argumentSchema
)
}
: []
)
let api = buildCompletionStreamAPI(
configuration.apiKey,
model,
url,
requestBody
)
#if DEBUG
Debugger.didSendRequestBody(body: requestBody)
#endif
return AsyncThrowingStream<StreamContent, Error> { continuation in
let task = Task {
do {
let (trunks, cancel) = try await api()
cancelTask = cancel
let proposedId = UUID().uuidString + String(Date().timeIntervalSince1970)
for try await trunk in trunks {
try Task.checkCancellation()
guard let delta = trunk.choices?.first?.delta else { continue }
// The api will always return a function call with JSON object.
// The first round will contain the function name and an empty argument.
// e.g. {"name":"weather","arguments":""}
// The other rounds will contain part of the arguments.
let functionCall = delta.function_call.map {
ChatMessage.FunctionCall(
name: $0.name ?? "",
arguments: $0.arguments ?? ""
)
}
await memory.streamMessage(
id: proposedId,
role: delta.role,
content: delta.content,
functionCall: functionCall
)
if let functionCall {
continuation.yield(.functionCall(functionCall))
}
if let content = delta.content {
continuation.yield(.text(content))
}
try await Task.sleep(nanoseconds: 3_000_000)
}
continuation.finish()
} catch let error as CancellationError {
continuation.finish(throwing: error)
} catch let error as NSError where error.code == NSURLErrorCancelled {
continuation.finish(throwing: error)
} catch {
await memory.appendMessage(.init(
role: .assistant,
content: error.localizedDescription
))
continuation.finish(throwing: error)
}
}
continuation.onTermination = { _ in
task.cancel()
}
}
}
/// Send the memory as prompt to ChatGPT, with stream disabled.
func sendMemoryAndWait() async throws -> ChatMessage? {
await memory.refresh()
guard let model = configuration.model else {
throw ChatGPTServiceError.chatModelNotAvailable
}
guard let url = URL(string: configuration.endpoint) else {
throw ChatGPTServiceError.endpointIncorrect
}
let messages = await memory.messages.map {
CompletionRequestBody.Message(
role: $0.role,
content: $0.content ?? "",
name: $0.name,
function_call: $0.functionCall.map {
.init(name: $0.name, arguments: $0.arguments)
}
)
}
let remainingTokens = await memory.remainingTokens
let requestBody = CompletionRequestBody(
model: model.info.modelName,
messages: messages,
temperature: configuration.temperature,
stream: true,
stop: configuration.stop.isEmpty ? nil : configuration.stop,
max_tokens: maxTokenForReply(
maxToken: model.info.maxTokens,
remainingTokens: remainingTokens
),
function_call: model.info.supportsFunctionCalling
? functionProvider.functionCallStrategy
: nil,
functions:
model.info.supportsFunctionCalling
? functionProvider.functions.map {
ChatGPTFunctionSchema(
name: $0.name,
description: $0.description,
parameters: $0.argumentSchema
)
}
: []
)
let api = buildCompletionAPI(
configuration.apiKey,
model,
url,
requestBody
)
#if DEBUG
Debugger.didSendRequestBody(body: requestBody)
#endif
let response = try await api()
guard let choice = response.choices.first else { return nil }
let message = ChatMessage(
id: response.id ?? UUID().uuidString,
role: choice.message.role,
content: choice.message.content,
name: choice.message.name,
functionCall: choice.message.function_call.map {
ChatMessage.FunctionCall(name: $0.name, arguments: $0.arguments ?? "")
}
)
await memory.appendMessage(message)
return message
}
/// When a function call is detected, but arguments are not yet ready, we can call this
/// to insert a message placeholder in memory.
func prepareFunctionCall(_ call: ChatMessage.FunctionCall, messageId: String) async {
guard let function = functionProvider.function(named: call.name) else { return }
let responseMessage = ChatMessage(
id: messageId,
role: .function,
content: nil,
name: call.name
)
await memory.appendMessage(responseMessage)
await function.prepare { [weak self] summary in
await self?.memory.updateMessage(id: messageId) { message in
message.summary = summary
}
}
}
/// Run a function call from the bot, and insert the result in memory.
@discardableResult
func runFunctionCall(
_ call: ChatMessage.FunctionCall,
messageId: String? = nil
) async -> String {
#if DEBUG
Debugger.didReceiveFunction(name: call.name, arguments: call.arguments)
#endif
let messageId = messageId ?? uuidGenerator()
guard let function = functionProvider.function(named: call.name) else {
return await fallbackFunctionCall(call, messageId: messageId)
}
// Insert the chat message into memory to indicate the start of the function.
let responseMessage = ChatMessage(
id: messageId,
role: .function,
content: nil,
name: call.name
)
await memory.appendMessage(responseMessage)
do {
// Run the function
let result = try await function.call(argumentsJsonString: call.arguments) {
[weak self] summary in
await self?.memory.updateMessage(id: messageId) { message in
message.summary = summary
}
}
#if DEBUG
Debugger.didReceiveFunctionResult(result: result.botReadableContent)
#endif
await memory.updateMessage(id: messageId) { message in
message.content = result.botReadableContent
}
return result.botReadableContent
} catch {
// For errors, use the error message as the result.
let content = "Error: \(error.localizedDescription)"
#if DEBUG
Debugger.didReceiveFunctionResult(result: content)
#endif
await memory.updateMessage(id: messageId) { message in
message.content = content
}
return content
}
}
/// Mock a function call result when the bot is calling a function that is not implemented.
func fallbackFunctionCall(
_ call: ChatMessage.FunctionCall,
messageId: String
) async -> String {
let memory = ConversationChatGPTMemory(systemPrompt: {
if call.name == "python" {
return """
Act like a Python interpreter.
I will give you Python code and you will execute it.
Reply with output of the code and tell me it's an answer generated by LLM.
"""
} else {
return """
You are a function simulator. Your name is \(call.name).
Act like a function.
I will send you the arguments.
Reply with output of the function and tell me it's an answer generated by LLM.
"""
}
}())
let service = ChatGPTService(
memory: memory,
configuration: OverridingChatGPTConfiguration(overriding: configuration, with: .init(
temperature: 0
)),
functionProvider: NoChatGPTFunctionProvider()
)
let content: String = await {
do {
return try await service.sendAndWait(content: """
\(call.arguments)
""") ?? "No result."
} catch {
return "No result."
}
}()
let responseMessage = ChatMessage(
id: messageId,
role: .function,
content: content,
name: call.name,
summary: "Finished running function."
)
await memory.appendMessage(responseMessage)
return content
}
}
extension ChatGPTService {
func changeBuildCompletionStreamAPI(_ builder: @escaping CompletionStreamAPIBuilder) {
buildCompletionStreamAPI = builder
}
func changeUUIDGenerator(_ generator: @escaping () -> String) {
uuidGenerator = generator
}
}
func maxTokenForReply(maxToken: Int, remainingTokens: Int?) -> Int? {
guard let remainingTokens else { return nil }
return min(maxToken / 2, remainingTokens)
}