File size: 15,607 Bytes
b50dda1
4ad8327
 
 
b50dda1
 
4ad8327
b50dda1
 
 
 
 
 
4ad8327
 
b50dda1
 
4ad8327
b50dda1
 
 
 
 
 
65ca8df
4452677
b50dda1
 
 
4ad8327
b50dda1
4ad8327
 
 
 
 
 
 
 
b50dda1
 
 
 
 
 
 
 
 
 
 
 
4ad8327
b50dda1
 
4ad8327
b50dda1
 
 
 
 
 
 
 
 
4ad8327
b50dda1
 
4ad8327
b50dda1
 
4ad8327
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b50dda1
 
 
 
4ad8327
 
b50dda1
 
 
4ad8327
 
 
 
 
 
 
 
 
 
 
b50dda1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4ad8327
 
 
 
 
b50dda1
 
 
 
 
 
 
 
4ad8327
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b50dda1
 
 
 
 
 
 
 
4ad8327
 
b50dda1
4ad8327
 
 
 
b50dda1
 
4ad8327
 
 
 
 
 
 
 
 
b50dda1
 
 
 
4ad8327
 
b50dda1
 
 
 
 
 
 
4ad8327
 
b50dda1
 
 
4ad8327
 
 
b50dda1
 
 
 
 
 
 
 
 
 
4ad8327
b50dda1
 
 
 
 
 
4ad8327
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b50dda1
 
4ad8327
b50dda1
4ad8327
 
b50dda1
4ad8327
 
 
 
 
 
 
 
 
 
 
 
 
b50dda1
4ad8327
 
 
 
 
 
 
 
b50dda1
4ad8327
 
 
 
 
 
b50dda1
 
 
 
4ad8327
b50dda1
4ad8327
 
b50dda1
4ad8327
 
 
b50dda1
4ad8327
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b50dda1
4ad8327
 
 
 
 
 
b50dda1
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
SAM-Z-1 Smart Load Balancing Cluster Head Node
- Light load: parallel gen/decode split for max speed
- Heavy load: 1 worker per request for throughput
"""

from fastapi import FastAPI, HTTPException
from fastapi.responses import StreamingResponse
from pydantic import BaseModel
import httpx
import asyncio
import json
import time
from typing import List, Optional, Dict
from collections import deque
import random

app = FastAPI(title="SAM-Z-1 Smart Cluster API", version="3.0.0")

# ============================================================================
# Configuration
# ============================================================================

WORKER_URLS = [
    "https://bc-ai-worker-2.hf.space",
    "https://bc-ai-worker-sam-z-api.hf.space",
]

HEALTH_CHECK_INTERVAL = 30
LOAD_CHECK_WINDOW = 10  # seconds to measure load

# Load thresholds
LIGHT_LOAD_THRESHOLD = 2  # requests in window
HEAVY_LOAD_THRESHOLD = 5  # requests in window

# Worker state
worker_health = {url: {"healthy": True, "last_check": 0, "active_requests": 0} for url in WORKER_URLS}
request_timestamps = deque(maxlen=100)  # track recent requests
current_load_mode = "light"  # "light" or "heavy"

# ============================================================================
# Request Models
# ============================================================================

class GenerateRequest(BaseModel):
    prompt: str
    max_tokens: int = 512
    temperature: float = 0.8
    top_k: int = 40
    top_p: float = 0.9
    repetition_penalty: float = 1.1
    stream: bool = True

class ChatMessage(BaseModel):
    role: str
    content: str

class ChatRequest(BaseModel):
    messages: List[ChatMessage]
    max_tokens: int = 512
    temperature: float = 0.8
    top_k: int = 40
    top_p: float = 0.9
    repetition_penalty: float = 1.1
    stream: bool = True

# ============================================================================
# Load Management
# ============================================================================

def get_current_load() -> int:
    """Calculate current load based on recent requests"""
    now = time.time()
    # Count requests in the last LOAD_CHECK_WINDOW seconds
    return sum(1 for ts in request_timestamps if now - ts < LOAD_CHECK_WINDOW)

def update_load_mode():
    """Update load mode based on current load"""
    global current_load_mode
    load = get_current_load()
    
    if load <= LIGHT_LOAD_THRESHOLD:
        current_load_mode = "light"
    elif load >= HEAVY_LOAD_THRESHOLD:
        current_load_mode = "heavy"
    # hysteresis zone between thresholds maintains current mode
    
    return current_load_mode, load

def track_request():
    """Track a new request"""
    request_timestamps.append(time.time())

def get_healthy_workers() -> List[str]:
    """Get list of healthy workers"""
    return [url for url, status in worker_health.items() if status["healthy"]]

def get_least_busy_worker() -> Optional[str]:
    """Get worker with fewest active requests"""
    healthy = get_healthy_workers()
    if not healthy:
        return None
    return min(healthy, key=lambda url: worker_health[url]["active_requests"])

def select_worker_pair() -> tuple:
    """Select 2 workers for parallel operation"""
    healthy = get_healthy_workers()
    if len(healthy) < 2:
        return (healthy[0], None) if len(healthy) == 1 else (None, None)
    
    # Sort by active requests, take 2 least busy
    sorted_workers = sorted(healthy, key=lambda url: worker_health[url]["active_requests"])
    return (sorted_workers[0], sorted_workers[1])

async def check_worker_health(worker_url: str) -> bool:
    """Check if a worker is healthy"""
    try:
        async with httpx.AsyncClient(timeout=5.0) as client:
            response = await client.get(f"{worker_url}/health")
            return response.status_code == 200
    except:
        return False

async def health_check_loop():
    """Background task to check worker health"""
    while True:
        for worker_url in WORKER_URLS:
            healthy = await check_worker_health(worker_url)
            worker_health[worker_url]["healthy"] = healthy
            worker_health[worker_url]["last_check"] = time.time()
            
            status = "βœ…" if healthy else "❌"
            active = worker_health[worker_url]["active_requests"]
            print(f"{status} {worker_url}: {'healthy' if healthy else 'unhealthy'} | Active: {active}")
        
        mode, load = update_load_mode()
        print(f"πŸ“Š Load mode: {mode.upper()} | Current load: {load} req/{LOAD_CHECK_WINDOW}s")
        
        await asyncio.sleep(HEALTH_CHECK_INTERVAL)

@app.on_event("startup")
async def startup_event():
    """Start health check loop on startup"""
    asyncio.create_task(health_check_loop())

# ============================================================================
# Generation Strategies
# ============================================================================

async def light_load_generation(
    generator_url: str,
    decoder_url: str,
    request_data: dict,
    endpoint: str = "generate"
):
    """
    LIGHT LOAD MODE: Split generation and decoding
    - Generator worker: produces token IDs only
    - Decoder worker: decodes token IDs to text
    This parallelizes the bottleneck!
    """
    
    # Queues for pipeline
    token_queue = asyncio.Queue(maxsize=10)
    text_queue = asyncio.Queue(maxsize=10)
    
    async def generate_tokens():
        """Worker 1: Generate token IDs"""
        try:
            worker_health[generator_url]["active_requests"] += 1
            
            # Request token IDs only mode
            request_data_tokens = {**request_data, "return_token_ids": True}
            
            async with httpx.AsyncClient(timeout=300.0) as client:
                async with client.stream(
                    "POST",
                    f"{generator_url}/{endpoint}",
                    json=request_data_tokens
                ) as response:
                    async for chunk in response.aiter_text():
                        if chunk.strip() and chunk.startswith("data: "):
                            try:
                                data = json.loads(chunk[6:])
                                if "token_id" in data:
                                    await token_queue.put(data["token_id"])
                                elif "done" in data:
                                    await token_queue.put(None)  # Signal end
                                    break
                            except:
                                pass
        except Exception as e:
            print(f"❌ Generator error: {e}")
            await token_queue.put(None)
        finally:
            worker_health[generator_url]["active_requests"] -= 1
    
    async def decode_tokens():
        """Worker 2: Decode token IDs to text"""
        try:
            worker_health[decoder_url]["active_requests"] += 1
            
            batch = []
            batch_size = 5  # decode in small batches for speed
            
            while True:
                try:
                    token_id = await asyncio.wait_for(token_queue.get(), timeout=1.0)
                    
                    if token_id is None:
                        # Decode remaining batch
                        if batch:
                            async with httpx.AsyncClient(timeout=10.0) as client:
                                response = await client.post(
                                    f"{decoder_url}/decode",
                                    json={"token_ids": batch}
                                )
                                text = response.json()["text"]
                                await text_queue.put(("text", text))
                        
                        await text_queue.put(("done", None))
                        break
                    
                    batch.append(token_id)
                    
                    # Decode batch when full
                    if len(batch) >= batch_size:
                        async with httpx.AsyncClient(timeout=10.0) as client:
                            response = await client.post(
                                f"{decoder_url}/decode",
                                json={"token_ids": batch}
                            )
                            text = response.json()["text"]
                            await text_queue.put(("text", text))
                        
                        batch = []
                
                except asyncio.TimeoutError:
                    continue
        
        except Exception as e:
            print(f"❌ Decoder error: {e}")
            await text_queue.put(("done", None))
        finally:
            worker_health[decoder_url]["active_requests"] -= 1
    
    # Start both pipelines
    gen_task = asyncio.create_task(generate_tokens())
    dec_task = asyncio.create_task(decode_tokens())
    
    # Stream decoded text
    accumulated_text = ""
    try:
        while True:
            msg_type, data = await text_queue.get()
            
            if msg_type == "done":
                break
            
            if msg_type == "text":
                accumulated_text += data
                yield f"data: {json.dumps({'delta': data, 'text': accumulated_text})}\n\n"
    
    finally:
        await gen_task
        await dec_task

async def heavy_load_generation(
    worker_url: str,
    request_data: dict,
    endpoint: str = "generate"
):
    """
    HEAVY LOAD MODE: Single worker per request
    Standard streaming for max throughput
    """
    try:
        worker_health[worker_url]["active_requests"] += 1
        
        async with httpx.AsyncClient(timeout=300.0) as client:
            async with client.stream(
                "POST",
                f"{worker_url}/{endpoint}",
                json=request_data
            ) as response:
                async for chunk in response.aiter_text():
                    if chunk.strip():
                        yield chunk
    
    except Exception as e:
        yield f"data: {json.dumps({'error': str(e)})}\n\n"
    
    finally:
        worker_health[worker_url]["active_requests"] -= 1

# ============================================================================
# API Endpoints
# ============================================================================

@app.get("/")
async def root():
    """API info"""
    healthy_count = len(get_healthy_workers())
    mode, load = update_load_mode()
    
    return {
        "name": "SAM-Z-1 Smart Cluster API",
        "version": "3.0.0",
        "mode": mode,
        "current_load": load,
        "workers": len(WORKER_URLS),
        "healthy_workers": healthy_count,
        "features": [
            "smart_load_balancing",
            "parallel_gen_decode",
            "adaptive_routing"
        ],
        "load_strategy": {
            "light": "parallel gen/decode split for speed",
            "heavy": "1 worker per request for throughput"
        },
        "endpoints": {
            "generate": "/v1/generate",
            "chat": "/v1/chat",
            "health": "/health",
            "workers": "/workers",
            "stats": "/stats"
        }
    }

@app.get("/health")
async def health():
    """Health check endpoint"""
    healthy_count = len(get_healthy_workers())
    mode, load = update_load_mode()
    
    return {
        "status": "healthy" if healthy_count > 0 else "unhealthy",
        "workers_total": len(WORKER_URLS),
        "workers_healthy": healthy_count,
        "load_mode": mode,
        "current_load": load
    }

@app.get("/workers")
async def workers_status():
    """Get status of all workers"""
    return {
        "workers": [
            {
                "url": url,
                "healthy": status["healthy"],
                "active_requests": status["active_requests"],
                "last_check": status["last_check"]
            }
            for url, status in worker_health.items()
        ]
    }

@app.get("/stats")
async def stats():
    """Get cluster statistics"""
    mode, load = update_load_mode()
    
    return {
        "load_mode": mode,
        "current_load": load,
        "load_window_seconds": LOAD_CHECK_WINDOW,
        "thresholds": {
            "light": LIGHT_LOAD_THRESHOLD,
            "heavy": HEAVY_LOAD_THRESHOLD
        },
        "recent_requests": len(request_timestamps),
        "worker_stats": {
            url: {
                "healthy": status["healthy"],
                "active": status["active_requests"]
            }
            for url, status in worker_health.items()
        }
    }

@app.post("/v1/generate")
async def generate(request: GenerateRequest):
    """Generate text with smart load balancing"""
    
    track_request()
    mode, load = update_load_mode()
    
    healthy = get_healthy_workers()
    if not healthy:
        raise HTTPException(status_code=503, detail="No healthy workers available")
    
    request_data = {
        "prompt": request.prompt,
        "max_tokens": request.max_tokens,
        "temperature": request.temperature,
        "top_k": request.top_k,
        "top_p": request.top_p,
        "repetition_penalty": request.repetition_penalty,
        "stream": True
    }
    
    print(f"🎯 Mode: {mode.upper()} | Load: {load} | Request: generate")
    
    if mode == "light" and len(healthy) >= 2:
        # LIGHT LOAD: parallel gen/decode
        generator, decoder = select_worker_pair()
        return StreamingResponse(
            light_load_generation(generator, decoder, request_data, "generate"),
            media_type="text/event-stream"
        )
    else:
        # HEAVY LOAD: single worker
        worker = get_least_busy_worker()
        return StreamingResponse(
            heavy_load_generation(worker, request_data, "generate"),
            media_type="text/event-stream"
        )

@app.post("/v1/chat")
async def chat(request: ChatRequest):
    """Chat completion with smart load balancing"""
    
    track_request()
    mode, load = update_load_mode()
    
    healthy = get_healthy_workers()
    if not healthy:
        raise HTTPException(status_code=503, detail="No healthy workers available")
    
    request_data = {
        "messages": [{"role": m.role, "content": m.content} for m in request.messages],
        "max_tokens": request.max_tokens,
        "temperature": request.temperature,
        "top_k": request.top_k,
        "top_p": request.top_p,
        "repetition_penalty": request.repetition_penalty,
        "stream": True
    }
    
    print(f"🎯 Mode: {mode.upper()} | Load: {load} | Request: chat")
    
    if mode == "light" and len(healthy) >= 2:
        # LIGHT LOAD: parallel gen/decode
        generator, decoder = select_worker_pair()
        return StreamingResponse(
            light_load_generation(generator, decoder, request_data, "chat"),
            media_type="text/event-stream"
        )
    else:
        # HEAVY LOAD: single worker
        worker = get_least_busy_worker()
        return StreamingResponse(
            heavy_load_generation(worker, request_data, "chat"),
            media_type="text/event-stream"
        )

# ============================================================================
# Launch
# ============================================================================

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(
        app,
        host="0.0.0.0",
        port=7860,
        log_level="info"
    )