groq-performance-tuning by jeremylongshore

---
name: groq-performance-tuning
description: |
  Optimize Groq API performance with caching, batching, and connection pooling.
  Use when experiencing slow API responses, implementing caching strategies,
  or optimizing request throughput for Groq integrations.
  Trigger with phrases like "groq performance", "optimize groq",
  "groq latency", "groq caching", "groq slow", "groq batch".
allowed-tools: Read, Write, Edit
version: 1.0.0
license: MIT
author: Jeremy Longshore <jeremy@intentsolutions.io>
compatible-with: claude-code, codex, openclaw
---
# Groq Performance Tuning

## Overview
Maximize Groq's ultra-low-latency LPU inference. Groq delivers sub-100ms token generation; tuning focuses on streaming efficiency, prompt caching, model selection for speed vs quality, and parallel request orchestration.

## Prerequisites
- Groq API key with rate limit awareness
- `groq-sdk` npm package installed
- Understanding of LLM token economics
- Monitoring for TTFT (time to first token)

## Instructions

### Step 1: Select Optimal Model for Speed
```typescript
import Groq from 'groq-sdk';

const groq = new Groq({ apiKey: process.env.GROQ_API_KEY });

// Model speed tiers (approximate TTFT):
// llama-3.3-70b-versatile: ~200ms TTFT, best quality
// llama-3.1-8b-instant:    ~80ms TTFT, fastest
// mixtral-8x7b-32768:      ~150ms TTFT, long context  # 32768 = configured value

async function fastCompletion(prompt: string) {
  return groq.chat.completions.create({
    model: 'llama-3.1-8b-instant', // Fastest model
    messages: [{ role: 'user', content: prompt }],
    temperature: 0,       // Deterministic = cacheable
    max_tokens: 256,      // Limit output for speed  # 256 bytes
  });
}
```

### Step 2: Streaming for Perceived Performance
```typescript
async function streamCompletion(
  messages: any[],
  onToken: (token: string) => void
) {
  const stream = await groq.chat.completions.create({
    model: 'llama-3.3-70b-versatile',
    messages,
    stream: true,
    max_tokens: 1024,  # 1024: 1 KB
  });

  let fullResponse = '';
  for await (const chunk of stream) {
    const token = chunk.choices[0]?.delta?.content || '';
    fullResponse += token;
    onToken(token);
  }
  return fullResponse;
}
```

### Step 3: Semantic Prompt Cache
```typescript
import { LRUCache } from 'lru-cache';
import { createHash } from 'crypto';

const promptCache = new LRUCache<string, string>({
  max: 500,  # HTTP 500 Internal Server Error
  ttl: 1000 * 60 * 10, // 10 min for deterministic prompts  # 1000: 1 second in ms
});

function hashPrompt(messages: any[], model: string): string {
  return createHash('sha256')
    .update(JSON.stringify({ messages, model }))
    .digest('hex');
}

async function cachedCompletion(messages: any[], model: string) {
  const key = hashPrompt(messages, model);
  const cached = promptCache.get(key);
  if (cached) return cached;

  const response = await groq.chat.completions.create({
    model,
    messages,
    temperature: 0,
  });

  const result = response.choices[0].message.content!;
  promptCache.set(key, result);
  return result;
}
```

### Step 4: Parallel Request Orchestration
```typescript
async function parallelCompletions(
  prompts: string[],
  concurrency = 5
) {
  const results: string[] = [];

  for (let i = 0; i < prompts.length; i += concurrency) {
    const batch = prompts.slice(i, i + concurrency);
    const batchResults = await Promise.all(
      batch.map(prompt =>
        cachedCompletion(
          [{ role: 'user', content: prompt }],
          'llama-3.1-8b-instant'
        )
      )
    );
    results.push(...batchResults);
  }
  return results;
}
```

## Error Handling
| Issue | Cause | Solution |
|-------|-------|----------|
| Rate limit 429 | Over RPM/TPM quota | Use exponential backoff, batch requests |
| High TTFT | Using 70b model | Switch to 8b-instant for latency-sensitive tasks |
| Stream disconnect | Network timeout | Implement reconnection with partial response recovery |
| Token overflow | max_tokens too high | Set conservative limits, truncate prompts |

## Examples

### Latency Benchmark
```typescript
async function benchmarkModels(prompt: string) {
  const models = ['llama-3.1-8b-instant', 'llama-3.3-70b-versatile'];

  for (const model of models) {
    const start = performance.now();
    await groq.chat.completions.create({
      model,
      messages: [{ role: 'user', content: prompt }],
      max_tokens: 100,
    });
    console.log(`${model}: ${(performance.now() - start).toFixed(0)}ms`);
  }
}
```

## Resources
- [Groq API Documentation](https://console.groq.com/docs)
- [Groq Rate Limits](https://console.groq.com/docs/rate-limits)
- [Groq Model Cards](https://console.groq.com/docs/models)

## Output

- Configuration files or code changes applied to the project
- Validation report confirming correct implementation
- Summary of changes made and their rationale