File size: 6,982 Bytes
fcc38f9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import asyncio
import random
import time
from typing import List, Dict, Any
from together import AsyncTogether

class RateLimiter:
    """Rate limiter for Together AI API"""
    def __init__(self, qps_limit=70, rpm_limit=4000, tps_limit=15000):
        self.qps_limit = qps_limit
        self.rpm_limit = rpm_limit
        self.tps_limit = tps_limit
        self.request_timestamps = []
        self.token_counts = []
        self.semaphore = asyncio.Semaphore(qps_limit)
    
    async def wait_if_needed(self, estimated_tokens=500):
        now = time.time()
        # Clean up old timestamps (older than 60 seconds)
        self.request_timestamps = [ts for ts in self.request_timestamps if now - ts < 60]
        self.token_counts = self.token_counts[-len(self.request_timestamps):]
        
        # Check RPM limit
        rpm_current = len(self.request_timestamps)
        if rpm_current >= self.rpm_limit:
            oldest = self.request_timestamps[0]
            wait_time = 60 - (now - oldest)
            if wait_time > 0:
                await asyncio.sleep(wait_time)
                return await self.wait_if_needed(estimated_tokens)
        
        # Check QPS limit (last 1 second)
        recent_requests = sum(1 for ts in self.request_timestamps if now - ts < 1)
        if recent_requests >= self.qps_limit:
            await asyncio.sleep(0.1)
            return await self.wait_if_needed(estimated_tokens)
        
        # Check TPS limit (last 1 second)
        recent_tokens = sum(tokens for ts, tokens in zip(self.request_timestamps, self.token_counts) if now - ts < 1)
        if recent_tokens + estimated_tokens >= self.tps_limit:
            await asyncio.sleep(0.2)
            return await self.wait_if_needed(estimated_tokens)
        
        # Update tracking
        self.request_timestamps.append(now)
        self.token_counts.append(estimated_tokens)

def run_together_request(prompts: List[str], config) -> List[Dict[str, Any]]:
    """
    Process prompts with Together AI, handling rate limits.
    
    Args:
        prompts: List of prompt strings to process
        config: Config object that supports config.get() method
    
    Returns:
        List of dictionaries with results for each prompt
    """
    # Process in batches if needed
    batch_size = config.get("batch_size", 20)
    if len(prompts) <= batch_size:
        return _process_batch(prompts, config)
    
    # For larger sets, process in batches
    all_results = []
    batches = [prompts[i:i + batch_size] for i in range(0, len(prompts), batch_size)]
    
    for i, batch in enumerate(batches):
        if len(batches) > 1:
            print(f"Processing batch {i+1}/{len(batches)} ({len(batch)} prompts)")
        batch_results = _process_batch(batch, config)
        all_results.extend(batch_results)
        if i < len(batches) - 1:
            time.sleep(0.5)
    
    return all_results

def _process_batch(prompts: List[str], config) -> List[Dict[str, Any]]:
    """Process a single batch with rate limiting"""
    async def _async_batch_completions():
        async_client = AsyncTogether()
        rate_limiter = RateLimiter(
            qps_limit=config.get("qps_limit", 70),
            rpm_limit=config.get("rpm_limit", 4000),
            tps_limit=config.get("tps_limit", 15000)
        )
        
        results = [{"response": "", "success": False, "retries": 0, "error": None} for _ in prompts]
        
        async def process_prompt(prompt: str, index: int) -> None:
            retries = 0
            # Estimate tokens (1 token ≈ 4 chars)
            estimated_prompt_tokens = len(prompt) // 4
            estimated_completion_tokens = config.get("max_tokens", 500)
            total_estimated_tokens = estimated_prompt_tokens + estimated_completion_tokens
            
            while retries <= config.get("max_retries", 3):
                try:
                    async with rate_limiter.semaphore:
                        await rate_limiter.wait_if_needed(total_estimated_tokens)
                        
                        response = await async_client.completions.create(
                            model=config.get("name", "meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8"),
                            prompt=prompt,
                            temperature=config.get("temperature", 0.1),
                            max_tokens=estimated_completion_tokens
                        )
                        
                        results[index] = {
                            "response": response.choices[0].text,
                            "success": True,
                            "retries": retries,
                            "error": None
                        }
                        return
                except Exception as e:
                    error_str = str(e)
                    retries += 1
                    
                    # Exponential backoff for rate limit errors
                    if "rate_limit" in error_str.lower():
                        backoff_time = config.get("retry_delay", 1) * (2 ** (retries - 1))
                        backoff_time += random.uniform(0, 1)  # Add jitter
                        backoff_time = min(backoff_time, 30)  # Cap at 30s
                        await asyncio.sleep(backoff_time)
                    elif retries <= config.get("max_retries", 3):
                        await asyncio.sleep(config.get("retry_delay", 1))
                    else:
                        results[index] = {
                            "response": f"Error after {retries} attempts",
                            "success": False,
                            "retries": retries,
                            "error": error_str
                        }
                        return
        
        tasks = [process_prompt(prompt, i) for i, prompt in enumerate(prompts)]
        
        try:
            await asyncio.wait_for(
                asyncio.gather(*tasks, return_exceptions=True), 
                timeout=config.get("request_timeout", 120)
            )
        except asyncio.TimeoutError:
            pass
        
        return results
    
    try:
        try:
            loop = asyncio.get_event_loop()
            if loop.is_running():
                loop = asyncio.new_event_loop()
                asyncio.set_event_loop(loop)
                results = loop.run_until_complete(_async_batch_completions())
                loop.close()
            else:
                results = loop.run_until_complete(_async_batch_completions())
        except RuntimeError:
            loop = asyncio.new_event_loop()
            asyncio.set_event_loop(loop)
            results = loop.run_until_complete(_async_batch_completions())
            loop.close()
    except Exception as e:
        return [{"response": f"Global error: {str(e)}", "success": False, "retries": 0, "error": str(e)} 
                for _ in prompts]
    
    return results