File size: 25,257 Bytes
5b47808
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e7b8ee2
5b47808
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
628112f
5b47808
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e7b8ee2
5b47808
e7b8ee2
 
 
 
 
 
5b47808
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e830d82
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5b47808
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e7b8ee2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5b47808
 
 
 
 
 
 
 
 
e7b8ee2
5b47808
 
 
 
e7b8ee2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5b47808
 
 
 
 
 
 
e7b8ee2
5b47808
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e7b8ee2
 
 
 
 
 
 
5b47808
e7b8ee2
 
 
5b47808
 
 
 
 
 
b6a0c21
 
5b47808
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e7b8ee2
 
 
 
 
 
5b47808
e7b8ee2
 
5b47808
 
 
 
 
 
 
 
 
 
 
 
 
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
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
# api_training_bridge.py
# ================================================================
# PONTE DE TREINAMENTO VIA API - AKIRA IA V21
# ================================================================
# Conecta os provedores API (Cerebras, OpenRouter, Mistral, etc.)
# ao sistema de treinamento, permitindo que todos os 5 níveis
# usem APIs para distilar conhecimento.
# ================================================================

import os
import time
import json
from typing import Optional, List, Dict, Any, Tuple
from loguru import logger

try:
    from . import config
except (ImportError, SystemError):
    try:
        import modules.config as config
    except ImportError:
        config = None


class APITrainingBridge:
    """
    Ponte entre provedores API e sistema de treinamento.
    
    Cria clientes API leves para uso exclusivo de treinamento,
    reutilizando as mesmas chaves de API e rotações do LLMManager.
    
    Uso:
        bridge = APITrainingBridge()
        resposta, provider = bridge.generate("Analise a emoção desta mensagem: ...")
    """

    def __init__(self):
        self._clients = {}
        self._providers = []
        self._rotation = {}
        self._initialized = False
        self._stats = {}
        self._init_clients()

    def _init_clients(self):
        """Inicializa clientes API leves para treinamento."""
        if self._initialized:
            return

        # 1. Cerebras (OpenAI SDK compat)
        self._setup_cerebras()

        # 2. OpenRouter (OpenAI SDK compat)
        self._setup_openrouter()

        # 3. Mistral (requests direto)
        self._setup_mistral()

        # 4. Groq (Groq SDK)
        self._setup_groq()

        # 5. Gemini (Google GenAI)
        self._setup_gemini()

        # 6. Cohere
        self._setup_cohere()

        # 7. HF Inference
        self._setup_hf_inference()

        self._initialized = True

        if self._providers:
            logger.info(
                f"🔗 [API-TRAINING] Bridge inicializado: "
                f"{', '.join(self._providers)} ({len(self._providers)} providers)"
            )
        else:
            logger.warning("🔗 [API-TRAINING] Nenhum provider disponível para treinamento")

    def _setup_cerebras(self):
        """Cerebras via rotação multi-conta."""
        try:
            from .cerebras_rotation import get_cerebras_rotation
            rotation = get_cerebras_rotation()
            if rotation.account_names:
                import openai
                key = rotation.get_current_api_key()
                if key:
                    client = openai.OpenAI(
                        api_key=key,
                        base_url="https://api.cerebras.ai/v1",
                        timeout=30.0,
                        max_retries=0,
                    )
                    self._clients["cerebras"] = client
                    self._rotation["cerebras"] = rotation
                    self._providers.append("cerebras")
                    logger.debug(f"✅ [API-TRAINING] Cerebras OK ({rotation.get_current_account_name()})")
        except Exception as e:
            logger.debug(f"[API-TRAINING] Cerebras skip: {e}")

    def _setup_openrouter(self):
        """OpenRouter via rotação multi-conta."""
        try:
            from .openrouter_rotation import get_openrouter_rotation
            rotation = get_openrouter_rotation()
            if rotation.api_keys:
                import openai
                key = rotation.get_current_key()
                if key:
                    client = openai.OpenAI(
                        api_key=key,
                        base_url="https://openrouter.ai/api/v1",
                        timeout=30.0,
                        max_retries=0,
                    )
                    self._clients["openrouter"] = client
                    self._rotation["openrouter"] = rotation
                    self._providers.append("openrouter")
                    logger.debug(f"✅ [API-TRAINING] OpenRouter OK ({rotation.get_current_account_name()})")
        except Exception as e:
            logger.debug(f"[API-TRAINING] OpenRouter skip: {e}")

    def _setup_mistral(self):
        """Mistral via API key direta ou rotação."""
        try:
            from .mistral_rotation import get_mistral_rotation
            rotation = get_mistral_rotation(config)
            if rotation:
                self._rotation["mistral"] = rotation
                self._providers.append("mistral")
                logger.debug(f"✅ [API-TRAINING] Mistral OK ({rotation.get_current_account_name()})")
            elif config and getattr(config, "MISTRAL_API_KEY", ""):
                self._providers.append("mistral")
                logger.debug("✅ [API-TRAINING] Mistral OK (chave única)")
        except Exception as e:
            logger.debug(f"[API-TRAINING] Mistral skip: {e}")

    def _setup_groq(self):
        """Groq SDK."""
        try:
            api_key = getattr(config, "GROQ_API_KEY", "") if config else ""
            if api_key and len(api_key) > 5:
                from groq import Groq
                client = Groq(api_key=api_key)
                self._clients["groq"] = client
                self._providers.append("groq")
                logger.debug("✅ [API-TRAINING] Groq OK")
        except Exception as e:
            logger.debug(f"[API-TRAINING] Groq skip: {e}")

    def _setup_gemini(self):
        """Google Gemini via genai."""
        try:
            if config:
                gemini_key = getattr(config, "GEMINI_API_KEY", "") or getattr(config, "GOOGLE_API_KEY", "")
                if gemini_key and len(gemini_key) > 5:
                    import google.generativeai as genai
                    genai.configure(api_key=gemini_key)
                    model_name = getattr(config, "GEMINI_MODEL", "gemini-2.5-flash")
                    self._clients["gemini"] = genai.GenerativeModel(model_name)
                    self._providers.append("gemini")
                    logger.debug("✅ [API-TRAINING] Gemini OK")
        except Exception as e:
            logger.debug(f"[API-TRAINING] Gemini skip: {e}")

    def _setup_cohere(self):
        """Cohere SDK."""
        try:
            api_key = getattr(config, "COHERE_API_KEY", "") if config else ""
            if api_key and len(api_key) > 5:
                from cohere import Client
                client = Client(api_key=api_key)
                self._clients["cohere"] = client
                self._providers.append("cohere")
                logger.debug("✅ [API-TRAINING] Cohere OK")
        except Exception as e:
            logger.debug(f"[API-TRAINING] Cohere skip: {e}")

    def _setup_hf_inference(self):
        """HF Inference via rotação multi-conta."""
        try:
            from .hf_inference_rotation import get_hf_inference_rotation
            rotation = get_hf_inference_rotation()
            configured = [
                acc for acc in rotation.account_order
                if os.getenv(rotation.accounts[acc])
            ]
            if configured:
                from huggingface_hub import InferenceClient
                token = rotation.get_current_api_token()
                if token:
                    client = InferenceClient(token=token, timeout=30.0)
                    self._clients["hf_inference"] = client
                    self._rotation["hf_inference"] = rotation
                    self._providers.append("hf_inference")
                    logger.debug(f"✅ [API-TRAINING] HF Inference OK ({len(configured)} contas)")
        except Exception as e:
            logger.debug(f"[API-TRAINING] HF Inference skip: {e}")

    # ============================================================
    # INTERFACE PÚBLICA
    # ============================================================

    def generate(
        self,
        prompt: str,
        system: str = "Responda de forma clara e direta.",
        max_tokens: int = 2048,
        temperature: float = 0.7,
        preferred_provider: Optional[str] = None,
    ) -> Tuple[str, str]:
        """
        Gera resposta via API para treinamento.
        
        Returns:
            (resposta_texto, provider_name) ou ("", "") se todos falharem
        """
        providers = self._providers[:]
        if preferred_provider and preferred_provider in providers:
            providers.remove(preferred_provider)
            providers.insert(0, preferred_provider)

        for provider in providers:
            try:
                text = self._call_provider(provider, prompt, system, max_tokens, temperature)
                if text and len(text.strip()) > 5:
                    self._record_call(provider, True)
                    return text.strip(), provider
                self._record_call(provider, False)
            except Exception as e:
                self._record_call(provider, False)
                err_str = str(e)
                logger.debug(f"[API-TRAINING] {provider} falhou: {e}")
                # 429 (rate limit temporário) → rotaciona conta
                if "429" in err_str or "rate" in err_str.lower():
                    self._handle_rate_limit(provider)
                # 402 (quota/crédito esgotado) → já tratado em _call_cerebras;
                # para outros providers, tenta rotacionar também.
                elif "402" in err_str or "payment" in err_str.lower():
                    self._handle_rate_limit(provider)

        return "", ""

    def generate_batch(
        self,
        prompts: List[Dict[str, str]],
        max_concurrent: int = 1,
        delay_between: float = 1.0,
    ) -> List[Dict[str, Any]]:
        """
        Gera respostas para batch de prompts.
        
        Args:
            prompts: Lista de {"prompt": str, "system": str (opcional)}
            max_concurrent: Máximo de chamadas simultâneas
            delay_between: Delay entre chamadas (rate limiting)
        
        Returns:
            Lista de {"prompt": str, "response": str, "provider": str, "success": bool}
        """
        results = []
        for i, item in enumerate(prompts):
            prompt = item.get("prompt", "")
            system = item.get("system", "Responda de forma clara e direta.")
            
            text, provider = self.generate(prompt, system)
            
            results.append({
                "prompt": prompt,
                "response": text,
                "provider": provider,
                "success": bool(text),
            })
            
            if delay_between > 0 and i < len(prompts) - 1:
                time.sleep(delay_between)
        
        return results

    def get_best_provider(self) -> str:
        """Retorna o provider com melhor taxa de sucesso."""
        best = ""
        best_score = -1
        
        for provider, stats in self._stats.items():
            total = stats.get("success", 0) + stats.get("failure", 0)
            if total < 3:
                continue
            rate = stats["success"] / total
            if rate > best_score:
                best_score = rate
                best = provider
        
        return best or (self._providers[0] if self._providers else "")

    def get_stats(self) -> Dict[str, Any]:
        """Retorna estatísticas de uso."""
        return {
            "providers_available": self._providers[:],
            "stats": dict(self._stats),
        }

    def is_available(self) -> bool:
        """Retorna se há providers disponíveis."""
        return bool(self._providers)

    def get_mac_training_context(self) -> Dict[str, Any]:
        """
        Retorna contexto de treinamento com estado do drive MAC.

        Inclui informações de saúde, uso e performance do disco
        para contextualizar dados de treinamento em sistemas Apple.
        """
        import subprocess

        context = {
            "platform": "unknown",
            "drives": [],
            "timestamp": time.time(),
        }

        try:
            import platform as plat
            system = plat.system()

            if system == "Darwin":
                context["platform"] = "macos"
                context["drives"] = self._get_mac_drive_states()
            else:
                context["platform"] = system.lower()
                context["drives"] = self._get_generic_drive_states()
        except Exception as e:
            logger.debug(f"[API-TRAINING] MAC context error: {e}")

        return context

    def _get_mac_drive_states(self) -> List[Dict[str, Any]]:
        """Coleta estado dos drives em macOS via diskutil."""
        drives = []
        try:
            result = subprocess.run(
                ["diskutil", "list", "-plist"],
                capture_output=True, text=True, timeout=10
            )
            if result.returncode == 0 and result.stdout:
                plist_data = __import__("plistlib").loads(result.stdout.encode())
                for disk in plist_data.get("AllDisks", []):
                    info = plist_data.get("AllDiskInformation", {}).get(disk, {})
                    drives.append({
                        "name": disk,
                        "volume_name": info.get("VolumeName", ""),
                        "mount_point": info.get("MountPoint", ""),
                        "total_bytes": info.get("TotalSize", 0),
                        "free_bytes": info.get("FreeSpace", 0),
                        "filesystem": info.get("Filesystem", ""),
                        "protocol": info.get("BusProtocol", ""),
                    })
        except Exception as e:
            logger.debug(f"[API-TRAINING] diskutil error: {e}")

        return drives

    def _get_generic_drive_states(self) -> List[Dict[str, Any]]:
        """Coleta estado dos drives em sistemas genéricos (Windows/Linux)."""
        drives = []
        try:
            import shutil
            usage = shutil.disk_usage("/")
            drives.append({
                "name": "/",
                "total_bytes": usage.total,
                "free_bytes": usage.free,
                "used_bytes": usage.used,
            })
        except Exception as e:
            logger.debug(f"[API-TRAINING] disk usage error: {e}")

        return drives

    # ============================================================
    # CHAMADAS POR PROVIDER
    # ============================================================

    def _call_provider(
        self, provider: str, prompt: str, system: str,
        max_tokens: int, temperature: float
    ) -> Optional[str]:
        """Roteia chamada para o provider correto."""
        if provider == "cerebras":
            return self._call_cerebras(prompt, system, max_tokens, temperature)
        elif provider == "openrouter":
            return self._call_openrouter(prompt, system, max_tokens, temperature)
        elif provider == "mistral":
            return self._call_mistral(prompt, system, max_tokens, temperature)
        elif provider == "groq":
            return self._call_groq(prompt, system, max_tokens, temperature)
        elif provider == "gemini":
            return self._call_gemini(prompt, system, max_tokens, temperature)
        elif provider == "cohere":
            return self._call_cohere(prompt, system, max_tokens, temperature)
        elif provider == "hf_inference":
            return self._call_hf_inference(prompt, system, max_tokens, temperature)
        return None

    def _call_cerebras(self, prompt, system, max_tokens, temperature):
        client = self._clients.get("cerebras")
        if not client:
            return None
        # Modelo Cerebras serverless default: gpt-oss-120b (120B params, 65k contexto, production).
        # Sobrescreve via config.CEREBRAS_MODEL ou env CEREBRAS_MODEL.
        model = getattr(config, "CEREBRAS_MODEL", "gpt-oss-120b") if config else "gpt-oss-120b"
        try:
            response = client.chat.completions.create(
                model=model,
                messages=[
                    {"role": "system", "content": system},
                    {"role": "user", "content": prompt},
                ],
                max_tokens=max_tokens,
                temperature=temperature,
            )
            return response.choices[0].message.content if response.choices else None
        except Exception as e:
            err_str = str(e)
            # 402 Payment Required → quota esgotada (não é rate limit temporário).
            # Blacklist 24h da conta atual e rotaciona, igual ao api.py.
            if "402" in err_str or "payment_required" in err_str.lower():
                rotation = self._rotation.get("cerebras")
                if rotation and hasattr(rotation, "handle_quota_402"):
                    try:
                        current_name = rotation.get_current_account_name()
                        rotation.handle_quota_402(current_name)
                        # Atualiza o cliente para a próxima conta disponível
                        new_key = rotation.get_current_api_key()
                        if new_key and new_key != client.api_key:
                            import openai
                            self._clients["cerebras"] = openai.OpenAI(
                                api_key=new_key,
                                base_url="https://api.cerebras.ai/v1",
                                timeout=30.0,
                                max_retries=0,
                            )
                            logger.info(f"🔄 [API-TRAINING] Cerebras 402 → rotacionado para '{rotation.get_current_account_name()}'")
                    except Exception as rotate_e:
                        logger.debug(f"[API-TRAINING] Cerebras 402 rotate error: {rotate_e}")
            raise

    def _call_openrouter(self, prompt, system, max_tokens, temperature, rotation=None):
        rotation = rotation or self._rotation.get("openrouter")
        client = self._clients.get("openrouter")
        if not client:
            return None

        # Atualiza chave se tem rotação
        if rotation:
            key = rotation.get_current_key()
            if key:
                client.api_key = key

        model = getattr(config, "OPENROUTER_MODEL", "nvidia/nemotron-3-ultra-550b-a55b:free") if config else "nvidia/nemotron-3-ultra-550b-a55b:free"
        try:
            response = client.chat.completions.create(
                model=model,
                messages=[
                    {"role": "system", "content": system},
                    {"role": "user", "content": prompt},
                ],
                max_tokens=max_tokens,
                temperature=temperature,
            )
            return response.choices[0].message.content if response.choices else None
        except Exception as e:
            err_str = str(e)
            # 429 → rotaciona conta OpenRouter
            if "429" in err_str or "rate" in err_str.lower():
                if rotation and hasattr(rotation, "handle_429_error"):
                    try:
                        rotation.handle_429_error()
                        new_key = rotation.get_current_key()
                        if new_key:
                            client.api_key = new_key
                    except Exception:
                        pass
            raise

    def _call_mistral(self, prompt, system, max_tokens, temperature):
        import requests
        
        rotation = self._rotation.get("mistral")
        api_key = ""
        if rotation:
            api_key = rotation.get_current_key()
        elif config:
            api_key = getattr(config, "MISTRAL_API_KEY", "")
        
        if not api_key:
            return None
        
        model = getattr(config, "MISTRAL_MODEL", "mistral-large-latest") if config else "mistral-large-latest"
        response = requests.post(
            "https://api.mistral.ai/v1/chat/completions",
            headers={"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"},
            json={
                "model": model,
                "messages": [
                    {"role": "system", "content": system},
                    {"role": "user", "content": prompt},
                ],
                "max_tokens": max_tokens,
                "temperature": temperature,
            },
            timeout=30,
        )
        if response.status_code == 200:
            data = response.json()
            choices = data.get("choices", [])
            return choices[0]["message"]["content"] if choices else None
        elif response.status_code == 429:
            # Rate limit → rotaciona conta Mistral e levanta para o generate() tratar
            rotation = self._rotation.get("mistral")
            if rotation and hasattr(rotation, "handle_429_error"):
                try:
                    rotation.handle_429_error()
                except Exception:
                    pass
            raise Exception("429 Rate Limit")
        elif response.status_code == 402:
            # Payment Required → quota esgotada
            raise Exception("402 Payment Required")
        return None

    def _call_groq(self, prompt, system, max_tokens, temperature):
        client = self._clients.get("groq")
        if not client:
            return None
        if len(system or '') + len(prompt or '') > 7500:
            return None
        model = getattr(config, "GROQ_MODEL", "groq/compound") if config else "groq/compound"
        response = client.chat.completions.create(
            model=model,
            messages=[
                {"role": "system", "content": system},
                {"role": "user", "content": prompt},
            ],
            max_tokens=max_tokens,
            temperature=temperature,
        )
        return response.choices[0].message.content if response.choices else None

    def _call_gemini(self, prompt, system, max_tokens, temperature):
        model = self._clients.get("gemini")
        if not model:
            return None
        full_prompt = f"{system}\n\n{prompt}"
        response = model.generate_content(
            full_prompt,
            generation_config={
                "max_output_tokens": max_tokens,
                "temperature": temperature,
            },
        )
        return response.text if response.text else None

    def _call_cohere(self, prompt, system, max_tokens, temperature):
        client = self._clients.get("cohere")
        if not client:
            return None
        model = getattr(config, "COHERE_MODEL", "command-r-plus-08-2024") if config else "command-r-plus-08-2024"
        response = client.chat(
            model=model,
            message=prompt,
            preamble=system,
            max_tokens=max_tokens,
            temperature=temperature,
        )
        return response.text if response.text else None

    def _call_hf_inference(self, prompt, system, max_tokens, temperature):
        client = self._clients.get("hf_inference")
        if not client:
            return None
        rotation = self._rotation.get("hf_inference")
        model = getattr(config, "MISTRAL_MODEL_HF", "mistralai/Mistral-7B-Instruct-v0.2") if config else "mistralai/Mistral-7B-Instruct-v0.2"
        
        # Atualiza token se tem rotação
        if rotation:
            token = rotation.get_current_api_token()
            if token:
                client.token = token
        
        full_prompt = f"{system}\n\n{prompt}"
        response = client.text_generation(
            full_prompt,
            model=model,
            max_new_tokens=max_tokens,
            temperature=temperature,
        )
        return response if isinstance(response, str) else None

    # ============================================================
    # UTILITÁRIOS
    # ============================================================

    def _record_call(self, provider: str, success: bool):
        """Registra resultado de chamada."""
        if provider not in self._stats:
            self._stats[provider] = {"success": 0, "failure": 0}
        if success:
            self._stats[provider]["success"] += 1
        else:
            self._stats[provider]["failure"] += 1

    def _handle_rate_limit(self, provider: str):
        """Trata rate limit rotacionando para próxima conta."""
        rotation = self._rotation.get(provider)
        if not rotation:
            return
        # Diferentes rotações usam nomes diferentes de método para 429.
        # Cerebras: handle_rate_limit_error(); OpenRouter/Mistral: handle_429_error().
        handler = getattr(rotation, "handle_429_error", None) or getattr(rotation, "handle_rate_limit_error", None)
        if handler:
            try:
                handler()
                logger.info(f"🔄 [API-TRAINING] {provider} rotacionado após rate limit")
            except Exception:
                pass


# Singleton
_bridge_instance = None

def get_api_training_bridge() -> APITrainingBridge:
    """Obtém ou cria singleton do bridge."""
    global _bridge_instance
    if _bridge_instance is None:
        _bridge_instance = APITrainingBridge()
    return _bridge_instance