@ondeinference / onde
Code Commits Issues Pulls Cloud Agent CI/CD
development
copilot/assess-inference-engine-complexity copilot/fix-onde-npm-sdk-v1-1-2-release development feature/diffusion-vision main
dart 527 lines 16.5 KB
Raw
1 // This file is automatically generated, so please do not edit it.
2 // @generated by `flutter_rust_bridge`@ 2.12.0.
3
4 // ignore_for_file: type=lint, unused_import, unnecessary_import
5
6 // ignore_for_file: invalid_use_of_internal_member
7
8 import 'frb_generated.dart';
9 import 'package:flutter_rust_bridge/flutter_rust_bridge_for_generated.dart';
10 import 'package:freezed_annotation/freezed_annotation.dart' hide protected;
11 part 'api.freezed.dart';
12
13 // These function are ignored because they are on traits that is not defined in current crate (put an empty `#[frb]` on it to unignore): `assert_fields_are_eq`, `assert_fields_are_eq`, `clone`, `clone`, `clone`, `clone`, `clone`, `clone`, `clone`, `clone`, `clone`, `clone`, `eq`, `eq`, `fmt`, `fmt`, `fmt`, `fmt`, `fmt`, `fmt`, `fmt`, `fmt`, `fmt`, `fmt`, `fmt`, `fmt`, `from`, `from`, `from`, `from`, `from`, `from`, `from`, `from`, `from`, `from`, `from`, `from`, `from`, `from`, `from`, `from`
14
15 /// Return the platform-appropriate default `GgufModelConfig`.
16 ///
17 /// Selects the 1.5B model on iOS / tvOS / Android and the 3B model on
18 /// macOS / Windows / Linux.
19 GgufModelConfig defaultModelConfig() =>
20 RustLib.instance.api.crateApiDefaultModelConfig();
21
22 /// Build a generic `UqffModelConfig`.
23 UqffModelConfig uqffModelConfig({
24 required String modelId,
25 required List<String> files,
26 required String displayName,
27 required String approxMemory,
28 String? chatTemplate,
29 }) => RustLib.instance.api.crateApiUqffModelConfig(
30 modelId: modelId,
31 files: files,
32 displayName: displayName,
33 approxMemory: approxMemory,
34 chatTemplate: chatTemplate,
35 );
36
37 /// `GgufModelConfig` for Qwen 2.5 1.5B Instruct Q4_K_M (~941 MB).
38 GgufModelConfig qwen2515BConfig() =>
39 RustLib.instance.api.crateApiQwen2515BConfig();
40
41 /// `GgufModelConfig` for Qwen 2.5 3B Instruct Q4_K_M (~1.93 GB).
42 GgufModelConfig qwen253BConfig() =>
43 RustLib.instance.api.crateApiQwen253BConfig();
44
45 /// `GgufModelConfig` for Qwen 2.5 Coder 1.5B Instruct Q4_K_M (~941 MB).
46 GgufModelConfig qwen25Coder15BConfig() =>
47 RustLib.instance.api.crateApiQwen25Coder15BConfig();
48
49 /// `GgufModelConfig` for Qwen 2.5 Coder 3B Instruct Q4_K_M (~1.93 GB).
50 GgufModelConfig qwen25Coder3BConfig() =>
51 RustLib.instance.api.crateApiQwen25Coder3BConfig();
52
53 /// `GgufModelConfig` for Qwen 3 0.6B Q4_K_M (~0.5 GB).
54 GgufModelConfig qwen306BConfig() =>
55 RustLib.instance.api.crateApiQwen306BConfig();
56
57 /// `GgufModelConfig` for Qwen 3 1.7B Q4_K_M (~1.3 GB).
58 GgufModelConfig qwen317BConfig() =>
59 RustLib.instance.api.crateApiQwen317BConfig();
60
61 /// `GgufModelConfig` for Qwen 3 4B Q4_K_M (~2.7 GB).
62 GgufModelConfig qwen34BConfig() => RustLib.instance.api.crateApiQwen34BConfig();
63
64 /// `GgufModelConfig` for Qwen 3 8B Q4_K_M (~5 GB).
65 GgufModelConfig qwen38BConfig() => RustLib.instance.api.crateApiQwen38BConfig();
66
67 /// `GgufModelConfig` for Qwen 3 14B Q4_K_M (~8.4 GB).
68 GgufModelConfig qwen314BConfig() =>
69 RustLib.instance.api.crateApiQwen314BConfig();
70
71 /// `GgufModelConfig` for Qwen 3 32B Q4_K_M (~19.8 GB).
72 GgufModelConfig qwen332BConfig() =>
73 RustLib.instance.api.crateApiQwen332BConfig();
74
75 /// `GgufModelConfig` for Qwen 3 4B Instruct 2507 Q4_K_M (~2.5 GB).
76 GgufModelConfig qwen34BInstruct2507Config() =>
77 RustLib.instance.api.crateApiQwen34BInstruct2507Config();
78
79 /// `GgufModelConfig` for Qwen 3 4B Thinking 2507 Q4_K_M (~2.5 GB).
80 GgufModelConfig qwen34BThinking2507Config() =>
81 RustLib.instance.api.crateApiQwen34BThinking2507Config();
82
83 /// `GgufModelConfig` for Qwen 3 30B-A3B Instruct 2507 (MoE) Q4_K_M (~18.6 GB).
84 GgufModelConfig qwen330BA3BInstruct2507Config() =>
85 RustLib.instance.api.crateApiQwen330BA3BInstruct2507Config();
86
87 /// Default sampling config: `temperature=0.7`, `top_p=0.95`, `max_tokens=512`.
88 SamplingConfig defaultSamplingConfig() =>
89 RustLib.instance.api.crateApiDefaultSamplingConfig();
90
91 /// Deterministic sampling config: `temperature=0.0`, greedy, `max_tokens=512`.
92 SamplingConfig deterministicSamplingConfig() =>
93 RustLib.instance.api.crateApiDeterministicSamplingConfig();
94
95 /// Mobile sampling config: `temperature=0.7`, `top_p=0.95`, `max_tokens=128`.
96 SamplingConfig mobileSamplingConfig() =>
97 RustLib.instance.api.crateApiMobileSamplingConfig();
98
99 /// Seed the HuggingFace cache environment for sandboxed platforms.
100 ///
101 /// On iOS (and Android), the default `~/.cache/huggingface/hub` path is outside
102 /// the app sandbox and writes fail with `Operation not permitted (os error 1)`.
103 ///
104 /// Call this **once** at startup — before any `load_gguf_model` or
105 /// `download_model` — with the app's writable data directory.
106 ///
107 /// # Dart usage
108 ///
109 /// ```dart
110 /// import 'package:path_provider/path_provider.dart';
111 ///
112 /// final dir = await getApplicationSupportDirectory();
113 /// configureCacheDir(appDataDir: dir.path);
114 /// ```
115 ///
116 /// Internally this:
117 /// 1. Creates `<app_data_dir>/huggingface/hub/` if it doesn't exist.
118 /// 2. Sets the `HF_HOME` and `HF_HUB_CACHE` environment variables.
119 /// 3. Seeds `mistralrs_core::GLOBAL_HF_CACHE` so the model loader never
120 /// falls back to `Cache::default()`.
121 void configureCacheDir({required String appDataDir}) =>
122 RustLib.instance.api.crateApiConfigureCacheDir(appDataDir: appDataDir);
123
124 // Rust type: RustOpaqueMoi<flutter_rust_bridge::for_generated::RustAutoOpaqueInner<OndeChatEngine>>
125 abstract class OndeChatEngine implements RustOpaqueInterface {
126 /// Clear the conversation history while keeping the model loaded.
127 ///
128 /// Returns the number of turns that were removed.
129 Future<BigInt> clearHistory();
130
131 /// Clear the current system prompt.
132 Future<void> clearSystemPrompt();
133
134 /// One-shot generation from an explicit message list.
135 ///
136 /// Does **not** modify the conversation history.
137 Future<InferenceResult> generate({
138 required List<ChatMessage> messages,
139 SamplingConfig? sampling,
140 });
141
142 /// Returns a clone of the full conversation history.
143 Future<List<ChatMessage>> history();
144
145 /// Returns a point-in-time snapshot of the engine's state.
146 Future<EngineInfo> info();
147
148 /// Returns `true` if a model is currently loaded and ready.
149 Future<bool> isLoaded();
150
151 /// Load the model assigned to this Onde app from the dashboard.
152 ///
153 /// Returns the wall-clock seconds taken to load the model.
154 Future<double> loadAssignedModel({
155 required String appId,
156 required String appSecret,
157 String? systemPrompt,
158 SamplingConfig? sampling,
159 });
160
161 /// Load a GGUF model into the engine.
162 ///
163 /// Returns the load duration in seconds on success.
164 Future<double> loadGgufModel({
165 required GgufModelConfig config,
166 String? systemPrompt,
167 SamplingConfig? sampling,
168 });
169
170 /// Load a UQFF model into the engine.
171 ///
172 /// Returns the load duration in seconds on success.
173 Future<double> loadUqffModel({
174 required UqffModelConfig config,
175 String? systemPrompt,
176 SamplingConfig? sampling,
177 });
178
179 /// Create a new engine with no model loaded.
180 factory OndeChatEngine() => RustLib.instance.api.crateApiOndeChatEngineNew();
181
182 /// Append a message to history without running inference.
183 ///
184 /// Useful for restoring a saved conversation or injecting context.
185 Future<void> pushHistory({required ChatMessage message});
186
187 /// Send a user message and receive a complete assistant reply.
188 ///
189 /// Both the user message and assistant reply are appended to the
190 /// conversation history on success.
191 Future<InferenceResult> sendMessage({required String message});
192
193 /// Replace the sampling configuration used for all subsequent inference.
194 Future<void> setSampling({required SamplingConfig sampling});
195
196 /// Set or replace the system prompt.
197 ///
198 /// The prompt is prepended to every inference request and is **not**
199 /// stored in the conversation history.
200 Future<void> setSystemPrompt({required String prompt});
201
202 /// Stream a reply token-by-token.
203 ///
204 /// Each [`StreamChunk`] carries a `delta` string. The final chunk has
205 /// `done == true`. The `sink` is closed automatically when the stream
206 /// ends or if the receiver is dropped.
207 Stream<StreamChunk> streamMessage({required String message});
208
209 /// Unload the current model and release its memory.
210 ///
211 /// Returns the display name of the model that was unloaded, or `None`
212 /// if no model was loaded.
213 Future<String?> unloadModel();
214 }
215
216 /// A single message in a conversation.
217 class ChatMessage {
218 final ChatRole role;
219 final String content;
220
221 const ChatMessage({required this.role, required this.content});
222
223 @override
224 int get hashCode => role.hashCode ^ content.hashCode;
225
226 @override
227 bool operator ==(Object other) =>
228 identical(this, other) ||
229 other is ChatMessage &&
230 runtimeType == other.runtimeType &&
231 role == other.role &&
232 content == other.content;
233 }
234
235 /// Role of a participant in a chat conversation.
236 enum ChatRole { system, user, assistant }
237
238 /// A point-in-time snapshot of the engine's state.
239 class EngineInfo {
240 final EngineStatus status;
241
242 /// Display name of the currently loaded model, if any.
243 final String? modelName;
244
245 /// Approximate memory footprint of the loaded model, if any.
246 final String? approxMemory;
247
248 /// Number of conversation turns in the current history.
249 final BigInt historyLength;
250
251 const EngineInfo({
252 required this.status,
253 this.modelName,
254 this.approxMemory,
255 required this.historyLength,
256 });
257
258 @override
259 int get hashCode =>
260 status.hashCode ^
261 modelName.hashCode ^
262 approxMemory.hashCode ^
263 historyLength.hashCode;
264
265 @override
266 bool operator ==(Object other) =>
267 identical(this, other) ||
268 other is EngineInfo &&
269 runtimeType == other.runtimeType &&
270 status == other.status &&
271 modelName == other.modelName &&
272 approxMemory == other.approxMemory &&
273 historyLength == other.historyLength;
274 }
275
276 /// Lifecycle status of the inference engine.
277 enum EngineStatus { unloaded, loading, ready, generating, error }
278
279 /// Configuration for loading a pre-quantised GGUF model.
280 class GgufModelConfig {
281 /// HuggingFace repository ID, e.g. `"bartowski/Qwen2.5-1.5B-Instruct-GGUF"`.
282 final String modelId;
283
284 /// GGUF filename(s) within the repository.
285 final List<String> files;
286
287 /// Optional explicit tokenizer model ID (required on Android).
288 final String? tokModelId;
289
290 /// Human-friendly display name, e.g. `"Qwen 2.5 1.5B"`.
291 final String displayName;
292
293 /// Approximate memory footprint, e.g. `"~941 MB (GGUF Q4_K_M)"`.
294 final String approxMemory;
295
296 const GgufModelConfig({
297 required this.modelId,
298 required this.files,
299 this.tokModelId,
300 required this.displayName,
301 required this.approxMemory,
302 });
303
304 @override
305 int get hashCode =>
306 modelId.hashCode ^
307 files.hashCode ^
308 tokModelId.hashCode ^
309 displayName.hashCode ^
310 approxMemory.hashCode;
311
312 @override
313 bool operator ==(Object other) =>
314 identical(this, other) ||
315 other is GgufModelConfig &&
316 runtimeType == other.runtimeType &&
317 modelId == other.modelId &&
318 files == other.files &&
319 tokModelId == other.tokModelId &&
320 displayName == other.displayName &&
321 approxMemory == other.approxMemory;
322 }
323
324 /// The result of a completed inference request.
325 class InferenceResult {
326 /// The generated assistant reply text.
327 final String text;
328
329 /// Wall-clock inference duration in seconds.
330 final double durationSecs;
331
332 /// Human-readable duration string, e.g. `"4.5s"` or `"2m 3.1s"`.
333 final String durationDisplay;
334
335 /// Finish reason reported by the model, e.g. `"stop"` or `"length"`.
336 final String finishReason;
337
338 /// Tool calls requested by the model (empty if none).
339 final List<ToolCallInfo> toolCalls;
340
341 const InferenceResult({
342 required this.text,
343 required this.durationSecs,
344 required this.durationDisplay,
345 required this.finishReason,
346 required this.toolCalls,
347 });
348
349 @override
350 int get hashCode =>
351 text.hashCode ^
352 durationSecs.hashCode ^
353 durationDisplay.hashCode ^
354 finishReason.hashCode ^
355 toolCalls.hashCode;
356
357 @override
358 bool operator ==(Object other) =>
359 identical(this, other) ||
360 other is InferenceResult &&
361 runtimeType == other.runtimeType &&
362 text == other.text &&
363 durationSecs == other.durationSecs &&
364 durationDisplay == other.durationDisplay &&
365 finishReason == other.finishReason &&
366 toolCalls == other.toolCalls;
367 }
368
369 @freezed
370 sealed class OndeError with _$OndeError implements FrbException {
371 const OndeError._();
372
373 const factory OndeError.noModelLoaded() = OndeError_NoModelLoaded;
374 const factory OndeError.alreadyLoaded({required String modelName}) =
375 OndeError_AlreadyLoaded;
376 const factory OndeError.modelBuild({required String reason}) =
377 OndeError_ModelBuild;
378 const factory OndeError.inference({required String reason}) =
379 OndeError_Inference;
380 const factory OndeError.cancelled() = OndeError_Cancelled;
381 const factory OndeError.other({required String reason}) = OndeError_Other;
382 }
383
384 /// Sampling parameters for text generation. All fields are optional — `None`
385 /// means "use the engine default".
386 class SamplingConfig {
387 final double? temperature;
388 final double? topP;
389 final BigInt? topK;
390 final double? minP;
391 final BigInt? maxTokens;
392 final double? frequencyPenalty;
393 final double? presencePenalty;
394
395 const SamplingConfig({
396 this.temperature,
397 this.topP,
398 this.topK,
399 this.minP,
400 this.maxTokens,
401 this.frequencyPenalty,
402 this.presencePenalty,
403 });
404
405 @override
406 int get hashCode =>
407 temperature.hashCode ^
408 topP.hashCode ^
409 topK.hashCode ^
410 minP.hashCode ^
411 maxTokens.hashCode ^
412 frequencyPenalty.hashCode ^
413 presencePenalty.hashCode;
414
415 @override
416 bool operator ==(Object other) =>
417 identical(this, other) ||
418 other is SamplingConfig &&
419 runtimeType == other.runtimeType &&
420 temperature == other.temperature &&
421 topP == other.topP &&
422 topK == other.topK &&
423 minP == other.minP &&
424 maxTokens == other.maxTokens &&
425 frequencyPenalty == other.frequencyPenalty &&
426 presencePenalty == other.presencePenalty;
427 }
428
429 /// A single streaming token chunk emitted during inference.
430 class StreamChunk {
431 /// The token text delta (may be empty on the final chunk).
432 final String delta;
433
434 /// `true` when this is the last chunk in the stream.
435 final bool done;
436
437 /// Finish reason (set only on the final chunk).
438 final String? finishReason;
439
440 const StreamChunk({
441 required this.delta,
442 required this.done,
443 this.finishReason,
444 });
445
446 @override
447 int get hashCode => delta.hashCode ^ done.hashCode ^ finishReason.hashCode;
448
449 @override
450 bool operator ==(Object other) =>
451 identical(this, other) ||
452 other is StreamChunk &&
453 runtimeType == other.runtimeType &&
454 delta == other.delta &&
455 done == other.done &&
456 finishReason == other.finishReason;
457 }
458
459 /// Structured tool-call request emitted by the model.
460 class ToolCallInfo {
461 final String id;
462 final String functionName;
463 final String arguments;
464
465 const ToolCallInfo({
466 required this.id,
467 required this.functionName,
468 required this.arguments,
469 });
470
471 @override
472 int get hashCode => id.hashCode ^ functionName.hashCode ^ arguments.hashCode;
473
474 @override
475 bool operator ==(Object other) =>
476 identical(this, other) ||
477 other is ToolCallInfo &&
478 runtimeType == other.runtimeType &&
479 id == other.id &&
480 functionName == other.functionName &&
481 arguments == other.arguments;
482 }
483
484 /// Configuration for loading a pre-quantised UQFF model.
485 class UqffModelConfig {
486 /// Base HuggingFace repository ID or local model directory.
487 final String modelId;
488
489 /// UQFF shard filename(s), first shard, or shorthand such as `q4k` / `4`.
490 final List<String> files;
491
492 /// Human-friendly display name.
493 final String displayName;
494
495 /// Approximate memory footprint.
496 final String approxMemory;
497
498 /// Optional Jinja chat template override.
499 final String? chatTemplate;
500
501 const UqffModelConfig({
502 required this.modelId,
503 required this.files,
504 required this.displayName,
505 required this.approxMemory,
506 this.chatTemplate,
507 });
508
509 @override
510 int get hashCode =>
511 modelId.hashCode ^
512 files.hashCode ^
513 displayName.hashCode ^
514 approxMemory.hashCode ^
515 chatTemplate.hashCode;
516
517 @override
518 bool operator ==(Object other) =>
519 identical(this, other) ||
520 other is UqffModelConfig &&
521 runtimeType == other.runtimeType &&
522 modelId == other.modelId &&
523 files == other.files &&
524 displayName == other.displayName &&
525 approxMemory == other.approxMemory &&
526 chatTemplate == other.chatTemplate;
527 }