Implements the OQ-1 follow-up to #215 v0.2.0: replace the placeholder GemmaLlmService stub with a real flutter_gemma 0.16.5 backend driving Gemma 4 E2B (litert-community/gemma-4-E2B-it-litert-lm, 2.41GB). Highlights: - GemmaLlmService.load → FlutterGemma.initialize + installModel.fromFile + getActiveModel; idempotent + FileSystemException on missing file. - generateStructured uses Gemma 4 native function calling via createChat(tools: [Tool(...)], toolChoice: required). Stream parsed by collectFunctionCall — first FCR wins, ParallelFCR first-call wins, TextResponse/ThinkingResponse skipped, errors sanitized to prevent prompt leakage. - main.dart wires _LazyLlmService adapter that resolves to GemmaLlmService when ModelLifecycle reports ready, MockLlmService otherwise. - ai_providers.dart pins real model URL + SHA-256 (181938...39a63c). - F2 hardening: ModelLifecycle.purge wraps each delete + meta remove in try/catch so a single OS-level flake cannot block opt-out. - Android: INTERNET / FOREGROUND_SERVICE / POST_NOTIFICATIONS permissions + R8 proguard-rules.pro keeping MediaPipe / LiteRT / TFLite / protobuf JNI entry points (release builds otherwise crash on first inference). Design-First: fn-gemma_llm_service.md updated to v2 — §C (_appendSchemaInstruction) deprecated after reading flutter_gemma 0.16.5 source (Gemma 4 SDK injects tool declarations via template; prompt-side append would double-wrap). Tests: - 10 new unit tests for collectFunctionCall covering all 8 fn-spec cases + 2 ParallelFunctionCallResponse paths. - All 81 existing tests still pass. - flutter analyze: 0 issues. Refs #218 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
194 lines
5.9 KiB
Dart
194 lines
5.9 KiB
Dart
import 'dart:async';
|
|
|
|
import 'package:flutter_riverpod/flutter_riverpod.dart';
|
|
|
|
import '../data/ai/llm_service.dart';
|
|
import '../data/ai/model_lifecycle.dart';
|
|
import '../data/db/app_database.dart' as drift;
|
|
import '../domain/ai/frame_candidate.dart';
|
|
import '../domain/ai/suggest_frame.dart';
|
|
import '../domain/models/frame_pattern.dart';
|
|
import 'providers.dart';
|
|
|
|
/// Gemma 4 E2B instruction-tuned LiteRT-LM checkpoint (#218 OQ-1 resolved).
|
|
/// Hosted on HuggingFace `litert-community/gemma-4-E2B-it-litert-lm`.
|
|
/// File ≈ 2.41GB; SHA-256 pinned for integrity check.
|
|
///
|
|
/// Tests / placeholder builds may override `modelLifecycleProvider` with
|
|
/// fixture URLs. Production builds optionally inject a private mirror via
|
|
/// `--dart-define=GEMMA_MODEL_URL=...` (see main.dart).
|
|
const _kModelUrl =
|
|
'https://huggingface.co/litert-community/gemma-4-E2B-it-litert-lm/resolve/main/gemma-4-E2B-it.litertlm';
|
|
const _kModelSha256 =
|
|
'181938105e0eefd105961417e8da75903eacda102c4fce9ce90f50b97139a63c';
|
|
|
|
final modelLifecycleProvider = Provider<ModelLifecycle>((ref) {
|
|
return ModelLifecycle(
|
|
meta: ref.watch(metaDaoProvider),
|
|
config: ModelConfig(
|
|
url: Uri.parse(_kModelUrl),
|
|
expectedSha256: _kModelSha256,
|
|
),
|
|
);
|
|
});
|
|
|
|
/// Read-only opt-in state. Default OFF; persisted in `meta_kv`.
|
|
final aiSettingsProvider = FutureProvider<bool>((ref) async {
|
|
final meta = ref.watch(metaDaoProvider);
|
|
final v = await meta.find(AiMetaKeys.optIn);
|
|
return v == 'true';
|
|
});
|
|
|
|
/// Toggles opt-in. On opt-out, purges model file via [ModelLifecycle.purge].
|
|
/// On opt-in, kicks off `ModelDownloadController.start()` so AC2 (progress UI)
|
|
/// has a stream to subscribe to.
|
|
class AiSettingsController {
|
|
AiSettingsController(this.ref);
|
|
final Ref ref;
|
|
|
|
Future<int> setOptIn(bool value) async {
|
|
final meta = ref.read(metaDaoProvider);
|
|
if (value) {
|
|
await meta.put(AiMetaKeys.optIn, 'true');
|
|
ref.invalidate(aiSettingsProvider);
|
|
ref.invalidate(modelAvailabilityProvider);
|
|
// AC2: opt-in triggers download stream so Settings UI can render
|
|
// progress + pause/resume. Fire-and-forget; controller emits states.
|
|
ref.read(modelDownloadControllerProvider.notifier).start();
|
|
return 0;
|
|
}
|
|
// opt-out: cancel any in-flight download, then purge.
|
|
ref.read(modelDownloadControllerProvider.notifier).cancel();
|
|
final freed = await ref.read(modelLifecycleProvider).purge();
|
|
await meta.put(AiMetaKeys.optIn, 'false');
|
|
ref.invalidate(aiSettingsProvider);
|
|
ref.invalidate(modelAvailabilityProvider);
|
|
return freed;
|
|
}
|
|
}
|
|
|
|
final aiSettingsControllerProvider = Provider<AiSettingsController>((ref) {
|
|
return AiSettingsController(ref);
|
|
});
|
|
|
|
/// AC2: streams DownloadProgress + supports pause/resume/cancel.
|
|
/// State `null` means idle (no active subscription).
|
|
class ModelDownloadController extends StateNotifier<DownloadProgress?> {
|
|
ModelDownloadController(this.ref) : super(null);
|
|
final Ref ref;
|
|
StreamSubscription<DownloadProgress>? _sub;
|
|
|
|
void start() {
|
|
cancel();
|
|
final lc = ref.read(modelLifecycleProvider);
|
|
_sub = lc.download().listen(
|
|
(p) {
|
|
state = p;
|
|
if (p.state == DownloadState.completed) {
|
|
ref.invalidate(modelAvailabilityProvider);
|
|
}
|
|
},
|
|
onError: (Object e) {
|
|
state = DownloadProgress(
|
|
bytesReceived: state?.bytesReceived ?? 0,
|
|
totalBytes: state?.totalBytes ?? -1,
|
|
state: DownloadState.failed,
|
|
errorMessage: e.toString(),
|
|
);
|
|
},
|
|
onDone: () {
|
|
_sub = null;
|
|
},
|
|
);
|
|
}
|
|
|
|
/// Pauses by cancelling the subscription. .tmp file + meta_kv preserved so
|
|
/// `start()` resumes via HTTP Range header.
|
|
void pause() {
|
|
_sub?.cancel();
|
|
_sub = null;
|
|
final cur = state;
|
|
if (cur != null && cur.state != DownloadState.completed) {
|
|
state = DownloadProgress(
|
|
bytesReceived: cur.bytesReceived,
|
|
totalBytes: cur.totalBytes,
|
|
state: DownloadState.paused,
|
|
);
|
|
}
|
|
}
|
|
|
|
void resume() => start();
|
|
|
|
void cancel() {
|
|
_sub?.cancel();
|
|
_sub = null;
|
|
state = null;
|
|
}
|
|
|
|
@override
|
|
void dispose() {
|
|
_sub?.cancel();
|
|
super.dispose();
|
|
}
|
|
}
|
|
|
|
final modelDownloadControllerProvider =
|
|
StateNotifierProvider<ModelDownloadController, DownloadProgress?>(
|
|
(ref) => ModelDownloadController(ref),
|
|
);
|
|
|
|
final modelAvailabilityProvider =
|
|
FutureProvider<ModelAvailability>((ref) async {
|
|
final lc = ref.watch(modelLifecycleProvider);
|
|
return lc.checkAvailability();
|
|
});
|
|
|
|
/// Loads FramePatterns from DB and converts to domain models.
|
|
final framePatternsProvider = FutureProvider<List<FramePatternModel>>(
|
|
(ref) async {
|
|
final db = ref.watch(appDatabaseProvider);
|
|
final rows = await db.select(db.framePatterns).get();
|
|
return rows.map(_toDomain).toList(growable: false);
|
|
},
|
|
);
|
|
|
|
FramePatternModel _toDomain(drift.FramePattern r) => FramePatternModel(
|
|
id: r.id,
|
|
domain: r.domain,
|
|
avoidanceKeyword: r.avoidanceKeyword,
|
|
l0Example: r.l0Example,
|
|
l1SimpleReplace: r.l1SimpleReplace,
|
|
l2Suggestion: r.l2Suggestion,
|
|
l3Identity: r.l3Identity,
|
|
);
|
|
|
|
/// Singleton LLM service for the app. v1 starts unloaded; first
|
|
/// [suggestFrame] triggers `.load()` via the dialog. Override in tests with
|
|
/// `MockLlmService`.
|
|
final llmServiceProvider = Provider<LlmService>((ref) {
|
|
throw UnimplementedError(
|
|
'llmServiceProvider must be overridden (Mock in tests, '
|
|
'GemmaLlmService after OQ-1 in production).',
|
|
);
|
|
});
|
|
|
|
/// `family` param wraps a SuggestFrameInput. Loads model lazily before
|
|
/// calling suggestFrame.
|
|
final frameSuggestionsProvider = FutureProvider.autoDispose
|
|
.family<List<FrameCandidate>, SuggestFrameInput>((ref, input) async {
|
|
final llm = ref.watch(llmServiceProvider);
|
|
final patterns = await ref.watch(framePatternsProvider.future);
|
|
if (!llm.isLoaded) {
|
|
try {
|
|
await llm.load();
|
|
} catch (_) {
|
|
return const [];
|
|
}
|
|
}
|
|
return suggestFrame(
|
|
input,
|
|
llm: llm,
|
|
framePatterns: patterns,
|
|
);
|
|
});
|