File size: 29,701 Bytes
a4b70d9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from __future__ import annotations

import random
import json
import uuid
import asyncio


from ...typing import AsyncResult, Messages, MediaListType
from ...requests import StreamSession, StreamResponse, FormData, raise_for_status
from ...providers.response import JsonConversation, FinishReason
from ...tools.media import merge_media
from ...image import to_bytes, is_accepted_format
from ..base_provider import AsyncGeneratorProvider, ProviderModelMixin
from ..helper import get_last_user_message
from ...errors import ModelNotFoundError, ResponseError
from ... import debug

class LegacyLMArena(AsyncGeneratorProvider, ProviderModelMixin):
    label = "LMArena (Legacy)"
    url = "https://legacy.lmarena.ai"
    api_endpoint = "/queue/join?"
    
    working = False
    
    default_model = "chatgpt-4o-latest-20250326"
    models = []
    
    # Models from HAR data (manually added)
    har_models = [
        "chatgpt-4o-latest-20250326", "gemini-2.5-pro-preview-05-06", "o3-2025-04-16", 
        "o4-mini-2025-04-16", "qwen3-235b-a22b", "mistral-medium-2505", 
        "gemini-2.5-flash-preview-04-17", "gpt-4.1-2025-04-14", 
        "llama-4-maverick-03-26-experimental", "grok-3-preview-02-24", 
        "claude-3-7-sonnet-20250219", "claude-3-7-sonnet-20250219-thinking-32k", 
        "deepseek-v3-0324", "llama-4-maverick-17b-128e-instruct", 
        "llama-4-scout-17b-16e-instruct", "gpt-4.1-mini-2025-04-14", 
        "gpt-4.1-nano-2025-04-14"
    ]

    # Models from JS data (manually added)
    js_models = [
        "gemini-2.0-flash-001", "gemini-2.0-flash-lite-preview-02-05", 
        "gemma-3-27b-it", "gemma-3-12b-it", "gemma-3-4b-it", 
        "deepseek-r1", "claude-3-5-sonnet-20241022", "o3-mini"
    ]
    
    # Updated vision models list from JS data
    vision_models = [
        "gemini-2.5-pro-preview-05-06", "o3-2025-04-16", "o4-mini-2025-04-16", 
        "mistral-medium-2505", "gemini-2.5-flash-preview-04-17", "gpt-4.1-2025-04-14", 
        "claude-3-7-sonnet-20250219", "claude-3-7-sonnet-20250219-thinking-32k", 
        "llama-4-maverick-17b-128e-instruct", "llama-4-scout-17b-16e-instruct", 
        "gpt-4.1-mini-2025-04-14", "gpt-4.1-nano-2025-04-14", "gemini-2.0-flash-001", 
        "gemini-2.0-flash-lite-preview-02-05", "gemma-3-27b-it", "claude-3-5-sonnet-20241022", 
        "gpt-4o-mini-2024-07-18", "gpt-4o-2024-11-20", "gpt-4o-2024-08-06", 
        "gpt-4o-2024-05-13", "mistral-small-3.1-24b-instruct-2503", 
        "claude-3-5-sonnet-20240620", "amazon-nova-pro-v1.0", "amazon-nova-lite-v1.0", 
        "qwen2.5-vl-32b-instruct", "qwen2.5-vl-72b-instruct", "gemini-1.5-pro-002", 
        "gemini-1.5-flash-002", "gemini-1.5-flash-8b-001", "gemini-1.5-pro-001", 
        "gemini-1.5-flash-001", "pixtral-large-2411", "step-1o-vision-32k-highres", 
        "claude-3-haiku-20240307", "claude-3-sonnet-20240229", "claude-3-opus-20240229", 
        "qwen-vl-max-1119", "qwen-vl-max-0809", "reka-core-20240904", 
        "reka-flash-20240904", "c4ai-aya-vision-32b", "pixtral-12b-2409"
    ]
    
    model_aliases = {
        # Existing aliases
        "claude-3.7-sonnet": "claude-3-7-sonnet-20250219",
        "claude-3.7-sonnet-thinking": "claude-3-7-sonnet-20250219-thinking-32k",
        "gpt-4o": "chatgpt-4o-latest-20250326",
        "grok-3": ["early-grok-3", "grok-3-preview-02-24",],
        "gemini-2.0-flash-thinking": ["gemini-2.0-flash-thinking-exp-01-21", "gemini-2.0-flash-thinking-exp-1219",],
        "gemini-2.0-pro-exp": "gemini-2.0-pro-exp-02-05",
        "gemini-2.0-flash": "gemini-2.0-flash-001",
        "o1": "o1-2024-12-17",
        "qwen-2.5-max": "qwen2.5-max",
        "o3": "o3-2025-04-16",
        "o4-mini": "o4-mini-2025-04-16",
        "gemini-1.5-pro": "gemini-1.5-pro-002",
        "grok-2": "grok-2-2024-08-13",
        "claude-3.5-sonnet": "claude-3-5-sonnet-20241022",
        "qwen-2.5-plus": "qwen2.5-plus-1127",
        "gpt-4o-mini": "gpt-4o-mini-2024-07-18",
        "gemini-1.5-flash": "gemini-1.5-flash-002",
        "llama-3.1-405b": ["llama-3.1-405b-instruct-bf16", "llama-3.1-405b-instruct-fp8",],
        "nemotron-70b": "llama-3.1-nemotron-70b-instruct",
        "grok-2-mini": "grok-2-mini-2024-08-13",
        "qwen-2.5-72b": "qwen2.5-72b-instruct",
        "qwen-2.5-vl-32b": "qwen2.5-vl-32b-instruct",
        "qwen-2.5-vl-72b": "qwen2.5-vl-72b-instruct",
        "gpt-4-turbo": "gpt-4-turbo-2024-04-09",
        "llama-3.3-70b": "llama-3.3-70b-instruct",
        "nemotron-49b": "llama-3.3-nemotron-49b-super-v1",
        "mistral-large": "mistral-large-2411",
        "pixtral-large": "pixtral-large-2411",
        "gpt-4": "gpt-4-0613",
        "gpt-4.1": "gpt-4.1-2025-04-14",
        "gpt-4.1-mini": "gpt-4.1-mini-2025-04-14",
        "gpt-4.1-nano": "gpt-4.1-nano-2025-04-14",
        "llama-3.1-70b": "llama-3.1-70b-instruct",
        "nemotron-253b": "llama-3.1-nemotron-ultra-253b-v1",
        "claude-3-opus": "claude-3-opus-20240229",
        "tulu-3-70b": "llama-3.1-tulu-3-70b",
        "claude-3.5-haiku": "claude-3-5-haiku-20241022",
        "reka-core": "reka-core-20240904",
        "gemma-2-27b": "gemma-2-27b-it",
        "gemma-3-27b": "gemma-3-27b-it",
        "gemma-3-12b": "gemma-3-12b-it",
        "gemma-3-4b": "gemma-3-4b-it",
        "deepseek-v2": "deepseek-v2-api-0628",
        "qwen-2.5-coder-32b": "qwen2.5-coder-32b-instruct",
        "gemma-2-9b": ["gemma-2-9b-it-simpo", "gemma-2-9b-it",],
        "command-a": "command-a-03-2025",
        "nemotron-51b": "llama-3.1-nemotron-51b-instruct",
        "mistral-small-24b": "mistral-small-24b-instruct-2501",
        "mistral-small-3.1-24b": "mistral-small-3.1-24b-instruct-2503",
        "glm-4": "glm-4-0520",
        "llama-3-70b": "llama-3-70b-instruct",
        "llama-4-maverick": "llama-4-maverick-17b-128e-instruct",
        "llama-4-scout": "llama-4-scout-17b-16e-instruct",
        "reka-flash": "reka-flash-20240904",
        "phi-4": "phi-4",
        "claude-3-sonnet": "claude-3-sonnet-20240229",
        "qwen-2-72b": "qwen2-72b-instruct",
        "qwen-3-235b": "qwen3-235b-a22b",
        "qwen-3-30b": "qwen3-30b-a3b",
        "qwen-3-32b": "qwen3-32b",
        "tulu-3-8b": "llama-3.1-tulu-3-8b",
        "command-r": ["command-r-08-2024", "command-r",],
        "codestral": "codestral-2405",
        "claude-3-haiku": "claude-3-haiku-20240307",
        "llama-3.1-8b": "llama-3.1-8b-instruct",
        "qwen-1.5-110b": "qwen1.5-110b-chat",
        "qwq-32b": "qwq-32b-preview",
        "llama-3-8b": "llama-3-8b-instruct",
        "qwen-1.5-72b": "qwen1.5-72b-chat",
        "gemma-2-2b": "gemma-2-2b-it",
        "qwen-vl-max": ["qwen-vl-max-1119", "qwen-vl-max-0809"],
        "gemini-2.5-pro": "gemini-2.5-pro-preview-05-06",
        "gemini-2.5-flash": "gemini-2.5-flash-preview-04-17",
        "mixtral-8x22b": "mixtral-8x22b-instruct-v0.1",
        "qwen-1.5-32b": "qwen1.5-32b-chat",
        "qwen-1.5-14b": "qwen1.5-14b-chat",
        "qwen-1.5-7b": "qwen1.5-7b-chat",
        "qwen-1.5-4b": "qwen1.5-4b-chat",
        "mistral-next": "mistral-next",
        "phi-3-medium": "phi-3-medium-4k-instruct",
        "phi-3-small": "phi-3-small-8k-instruct",
        "phi-3-mini": ["phi-3-mini-4k-instruct-june-2024", "phi-3-mini-4k-instruct", "phi-3-mini-128k-instruct"],
        "tulu-2-70b": "tulu-2-dpo-70b",
        "llama-2-70b": ["llama-2-70b-chat", "llama2-70b-steerlm-chat"],
        "llama-2-13b": "llama-2-13b-chat",
        "llama-2-7b": "llama-2-7b-chat",
        "hermes-2-dpo": "nous-hermes-2-mixtral-8x7b-dpo",
        "pplx-7b-online":"pplx-7b-online",
        "deepseek-67b": "deepseek-llm-67b-chat",
        "openhermes-2.5-7b": "openhermes-2.5-mistral-7b",
        "mistral-7b": "mistral-7b-instruct-v0.2",
        "llama-3.2-3b": "llama-3.2-3b-instruct",
        "llama-3.2-1b": "llama-3.2-1b-instruct",
        "codellama-34b": "codellama-34b-instruct",
        "codellama-70b": "codellama-70b-instruct",
        "qwen-14b": "qwen-14b-chat",
        "gpt-3.5-turbo":  "gpt-3.5-turbo-1106",
        "mixtral-8x7b": "mixtral-8x7b-instruct-v0.1",
        "dbrx-instruct": "dbrx-instruct-preview",
    }

    @classmethod
    def get_models(cls):
        """Get models with improved fallback sources"""
        if cls.models:  # Return cached models if already loaded
            return cls.models
            
        try:
            # Try to fetch models from Google Storage first
            url = "https://storage.googleapis.com/public-arena-no-cors/p2l-explorer/data/overall/arena.json"
            import requests
            response = requests.get(url, timeout=5)
            response.raise_for_status()
            data = response.json()
            leaderboard_models = [model[0] for model in data.get("leaderboard", [])]
            
            # Combine models from all sources and remove duplicates
            all_models = list(set(leaderboard_models + cls.har_models + cls.js_models))
            
            if all_models:
                # Ensure default model is at index 0
                if cls.default_model in all_models:
                    all_models.remove(cls.default_model)
                all_models.insert(0, cls.default_model)
                cls.models = all_models
                return cls.models
        except Exception as e:
            # Log the error and fall back to alternative sources
            debug.log(f"Failed to fetch models from Google Storage: {str(e)}")
        
        # Fallback: Use combined har_models and js_models
        combined_models = list(set(cls.har_models + cls.js_models))
        if combined_models:
            if cls.default_model in combined_models:
                combined_models.remove(cls.default_model)
            combined_models.insert(0, cls.default_model)
            cls.models = combined_models
            return cls.models
        
        # Final fallback: Use vision_models
        models = cls.vision_models.copy()
        if cls.default_model not in models:
            models.insert(0, cls.default_model)
        cls.models = models
        
        return cls.models

    @classmethod
    def get_model(cls, model: str) -> str:
        """Get the internal model name from the user-provided model name."""
        if not model:
            return cls.default_model
        
        # Ensure models are loaded
        if not cls.models:
            cls.get_models()
        
        # Check if the model exists directly in our models list
        if model in cls.models:
            return model
        
        # Check if there's an alias for this model
        if model in cls.model_aliases:
            alias = cls.model_aliases[model]
            # If the alias is a list, randomly select one of the options
            if isinstance(alias, list):
                selected_model = random.choice(alias)
                debug.log(f"LegacyLMArena: Selected model '{selected_model}' from alias '{model}'")
                return selected_model
            debug.log(f"LegacyLMArena: Using model '{alias}' for alias '{model}'")
            return alias
        
        # If model still not found, check in all available model sources directly
        all_available_models = list(set(cls.har_models + cls.js_models + cls.vision_models))
        if model in all_available_models:
            return model
        
        raise ModelNotFoundError(f"LegacyLMArena: Model {model} not found")

    @classmethod
    def _build_payloads(cls, model_id: str, session_hash: str, text: str, files: list, max_tokens: int, temperature: float, top_p: float):
        """Build payloads for new conversations"""
        first_payload = {
            "data": [
                None,
                model_id,
                {"text": text, "files": files},
                {
                    "text_models": [model_id],
                    "all_text_models": [model_id],
                    "vision_models": [],
                    "all_vision_models": [],
                    "image_gen_models": [],
                    "all_image_gen_models": [],
                    "search_models": [],
                    "all_search_models": [],
                    "models": [model_id],
                    "all_models": [model_id],
                    "arena_type": "text-arena"
                }
            ],
            "event_data": None,
            "fn_index": 119,
            "trigger_id": 159,
            "session_hash": session_hash
        }

        second_payload = {
            "data": [],
            "event_data": None,
            "fn_index": 120,
            "trigger_id": 159,
            "session_hash": session_hash
        }

        third_payload = {
            "data": [None, temperature, top_p, max_tokens],
            "event_data": None,
            "fn_index": 121,
            "trigger_id": 159,
            "session_hash": session_hash
        }

        return first_payload, second_payload, third_payload

    @classmethod
    def _build_continuation_payloads(cls, model_id: str, session_hash: str, text: str, max_tokens: int, temperature: float, top_p: float):
        """Renamed from _build_second_payloads for clarity"""
        first_payload = {
            "data":[None,model_id,text,{
                "text_models":[model_id],
                "all_text_models":[model_id],
                "vision_models":[],
                "image_gen_models":[],
                "all_image_gen_models":[],
                "search_models":[],
                "all_search_models":[],
                "models":[model_id],
                "all_models":[model_id],
                "arena_type":"text-arena"}],
            "event_data": None,
            "fn_index": 122,
            "trigger_id": 157,
            "session_hash": session_hash
        }

        second_payload = {
            "data": [],
            "event_data": None,
            "fn_index": 123,
            "trigger_id": 157,
            "session_hash": session_hash
        }

        third_payload = {
            "data": [None, temperature, top_p, max_tokens],
            "event_data": None,
            "fn_index": 124,
            "trigger_id": 157,
            "session_hash": session_hash
        }

        return first_payload, second_payload, third_payload

    @classmethod
    async def create_async_generator(
        cls,
        model: str,
        messages: Messages,
        proxy: str = None,
        media: MediaListType = None,
        max_tokens: int = 4096,
        temperature: float = 0.7,
        top_p: float = 1,
        conversation: JsonConversation = None,
        return_conversation: bool = True,
        max_retries: int = 1,
        **kwargs
    ) -> AsyncResult:
        async def read_response(response: StreamResponse):
            returned_data = ""
            async for line in response.iter_lines():
                if not line:
                    continue
                    
                # Handle both "data: " prefix and raw JSON
                if line.startswith(b"data: "):
                    line = line[6:]
                
                # Skip empty lines or non-JSON data
                line = line.strip()
                if not line or line == b"[DONE]":
                    continue
                    
                try:
                    json_data = json.loads(line)
                    
                    # Process data based on message type
                    if json_data.get("msg") == "process_generating":
                        output_data = json_data.get("output", {}).get("data", [])
                        if len(output_data) > 1 and output_data[1]:
                            # Extract content from various response formats
                            data = output_data[1]
                            content = None
                            
                            if isinstance(data, list):
                                if data and data[0] == "replace" and len(data) > 2:
                                    content = data[2]
                                elif data and isinstance(data[0], list) and len(data[0]) > 2:
                                    content = data[0][2]
                            elif isinstance(data, str):
                                # Handle direct string responses
                                content = data
                            
                            if content:
                                # Clean up content
                                if isinstance(content, str):
                                    if content.endswith("β–Œ"):
                                        content = content[:-1]
                                    if content in ['<span class="cursor"></span> ', 'update', '']:
                                        continue
                                    if content.startswith(returned_data):
                                        content = content[len(returned_data):]
                                    if content:
                                        returned_data += content
                                        yield content
                    
                    # Process completed messages
                    elif json_data.get("msg") == "process_completed":
                        output_data = json_data.get("output", {}).get("data", [])
                        if len(output_data) > 1:
                            # Handle both list and direct content
                            if isinstance(output_data[1], list):
                                for item in output_data[1]:
                                    if isinstance(item, list) and len(item) > 1:
                                        content = item[1]
                                    elif isinstance(item, str):
                                        content = item
                                    else:
                                        continue
                                        
                                    if content and content != returned_data and content != '<span class="cursor"></span> ':
                                        if "**NETWORK ERROR DUE TO HIGH TRAFFIC." in content:
                                            raise ResponseError(content)
                                        if content.endswith("β–Œ"):
                                            content = content[:-1]
                                        new_content = content
                                        if content.startswith(returned_data):
                                            new_content = content[len(returned_data):]
                                        if new_content:
                                            returned_data = content
                                            yield new_content
                            elif isinstance(output_data[1], str) and output_data[1]:
                                # Direct string content
                                content = output_data[1]
                                if content != returned_data:
                                    if content.endswith("β–Œ"):
                                        content = content[:-1]
                                    new_content = content
                                    if content.startswith(returned_data):
                                        new_content = content[len(returned_data):]
                                    if new_content:
                                        returned_data = content
                                        yield new_content
                                        
                    # Also check for other message types that might contain content
                    elif json_data.get("msg") in ["process_starts", "heartbeat"]:
                        # These are status messages, skip them but don't error
                        continue
                        
                except json.JSONDecodeError:
                    # Skip non-JSON lines
                    continue
                except Exception as e:
                    if max_retries == 1:
                        raise e
                    debug.log(f"Error parsing response: {str(e)}")
                    continue
        
        # Get the actual model name
        model = cls.get_model(model)
        prompt = get_last_user_message(messages)
        
        async with StreamSession(impersonate="chrome") as session:
            # Add retry logic for better reliability
            retry_count = 0
            
            while retry_count < max_retries:
                try:
                    # Handle new conversation
                    if conversation is None:
                        conversation = JsonConversation(session_hash=str(uuid.uuid4()).replace("-", ""))
                        media_objects = []
                        
                        # Process media if present
                        media = list(merge_media(media, messages))
                        if media:
                            data = FormData()
                            for i in range(len(media)):
                                media[i] = (to_bytes(media[i][0]), media[i][1])
                            for image, image_name in media:
                                data.add_field(f"files", image, filename=image_name)
                            
                            # Upload media files
                            async with session.post(f"{cls.url}/upload", params={"upload_id": conversation.session_hash}, data=data) as response:
                                await raise_for_status(response)
                                image_files = await response.json()
                            
                            # Format media objects for API request
                            media_objects = [{
                                "path": image_file,
                                "url": f"{cls.url}/file={image_file}",
                                "orig_name": media[i][1],
                                "size": len(media[i][0]),
                                "mime_type": is_accepted_format(media[i][0]),
                                "meta": {
                                    "_type": "gradio.FileData"
                                }
                            } for i, image_file in enumerate(image_files)]
                        
                        # Build payloads for new conversation
                        first_payload, second_payload, third_payload = cls._build_payloads(
                            model, conversation.session_hash, prompt, media_objects, 
                            max_tokens, temperature, top_p
                        )
                        
                        headers = {
                            "Content-Type": "application/json",
                            "Accept": "application/json",
                        }
                        
                        # Send the three required requests with small delays
                        async with session.post(f"{cls.url}{cls.api_endpoint}", json=first_payload, proxy=proxy, headers=headers) as response:
                            await raise_for_status(response)
                        
                        # Small delay between requests
                        await asyncio.sleep(0.1)
                        
                        async with session.post(f"{cls.url}{cls.api_endpoint}", json=second_payload, proxy=proxy, headers=headers) as response:
                            await raise_for_status(response)
                        
                        await asyncio.sleep(0.1)
                        
                        async with session.post(f"{cls.url}{cls.api_endpoint}", json=third_payload, proxy=proxy, headers=headers) as response:
                            await raise_for_status(response)
                        
                        # Small delay before streaming
                        await asyncio.sleep(0.2)
                        
                        # Stream the response
                        stream_url = f"{cls.url}/queue/data?session_hash={conversation.session_hash}"
                        async with session.get(stream_url, headers={"Accept": "text/event-stream"}, proxy=proxy) as response:
                            await raise_for_status(response)
                            count = 0
                            has_content = False
                            
                            # Add timeout for response
                            try:
                                async with asyncio.timeout(30):  # 30 second timeout
                                    async for chunk in read_response(response):
                                        count += 1
                                        has_content = True
                                        yield chunk
                            except asyncio.TimeoutError:
                                if not has_content:
                                    raise RuntimeError("Response timeout - no data received from server")
                        
                        # Only raise error if we truly got no content
                        if count == 0 and not has_content:
                            retry_count += 1
                            if retry_count < max_retries:
                                debug.log(f"No response received, retrying... (attempt {retry_count + 1}/{max_retries})")
                                await asyncio.sleep(1)  # Wait before retry
                                conversation = None  # Reset conversation for retry
                                continue
                            else:
                                raise RuntimeError("No response from server after multiple attempts")
                        
                        # Success - break retry loop
                        break
                        
                    # Handle continuation of existing conversation
                    else:
                        # Build payloads for conversation continuation
                        first_payload, second_payload, third_payload = cls._build_continuation_payloads(
                            model, conversation.session_hash, prompt, max_tokens, temperature, top_p
                        )
                        
                        headers = {
                            "Content-Type": "application/json",
                            "Accept": "application/json",
                        }
                        
                        # Send the three required requests with delays
                        async with session.post(f"{cls.url}{cls.api_endpoint}", json=first_payload, proxy=proxy, headers=headers) as response:
                            await raise_for_status(response)
                        
                        await asyncio.sleep(0.1)
                        
                        async with session.post(f"{cls.url}{cls.api_endpoint}", json=second_payload, proxy=proxy, headers=headers) as response:
                            await raise_for_status(response)
                        
                        await asyncio.sleep(0.1)
                        
                        async with session.post(f"{cls.url}{cls.api_endpoint}", json=third_payload, proxy=proxy, headers=headers) as response:
                            await raise_for_status(response)
                        
                        await asyncio.sleep(0.2)
                        
                        # Stream the response
                        stream_url = f"{cls.url}/queue/data?session_hash={conversation.session_hash}"
                        async with session.get(stream_url, headers={"Accept": "text/event-stream"}, proxy=proxy) as response:
                            await raise_for_status(response)
                            count = 0
                            has_content = False
                            
                            try:
                                async with asyncio.timeout(30):
                                    async for chunk in read_response(response):
                                        count += 1
                                        has_content = True
                                        yield chunk
                            except asyncio.TimeoutError:
                                if not has_content:
                                    raise RuntimeError("Response timeout - no data received from server")
                        
                        if count == 0 and not has_content:
                            raise RuntimeError("No response from server in conversation continuation")
                        
                        # Success - break retry loop
                        break
                        
                except Exception as e:
                    if retry_count < max_retries - 1:
                        retry_count += 1
                        debug.log(f"Error occurred: {str(e)}, retrying... (attempt {retry_count + 1}/{max_retries})")
                        await asyncio.sleep(1)
                        conversation = None  # Reset for retry
                        continue
                    else:
                        raise
            
            # Return conversation object for future interactions
            if return_conversation and conversation:
                yield conversation
            
            # Yield finish reason if we hit token limit
            if count >= max_tokens:
                yield FinishReason("length")