Skip to content

armgit / ruflo

Public Imported from GitHub
0

🌊 The original agent harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, federation, vector RAG integration, and native Claude Code / Codex / Hermes and many more Integrated

checkpoint: File edit:

rUv683734a10c41
diff --git a/v3/@claude-flow/performance/src/attention-integration.ts b/v3/@claude-flow/performance/src/attention-integration.ts
index e55cccce56..849e09909a 100644
--- a/v3/@claude-flow/performance/src/attention-integration.ts
+++ b/v3/@claude-flow/performance/src/attention-integration.ts
@@ -163,23 +163,54 @@ export class FlashAttentionOptimizer {
}
}
- // Benchmark Flash Attention
- const flashStart = performance.now();
- for (let i = 0; i < iterations; i++) {
- this.flashAttention.computeRaw(query, keys, values);
- }
- const flashEnd = performance.now();
- const flashTimeMs = flashEnd - flashStart;
- const flashAvgMs = flashTimeMs / iterations;
+ // Force garbage collection if available for accurate memory measurement
+ this.forceGC();
+
+ // Measure baseline memory usage
+ const baselineMemoryBefore = this.getMemoryUsage();
+ let baselinePeakMemory = baselineMemoryBefore;
- // Benchmark baseline (DotProduct)
+ // Benchmark baseline (DotProduct) - run first to establish baseline memory
const baselineStart = performance.now();
for (let i = 0; i < iterations; i++) {
this.baselineAttention.computeRaw(query, keys, values);
+ // Sample memory periodically (every 100 iterations to reduce overhead)
+ if (i % 100 === 0) {
+ const currentMemory = this.getMemoryUsage();
+ if (currentMemory > baselinePeakMemory) {
+ baselinePeakMemory = currentMemory;
+ }
+ }
}
const baselineEnd = performance.now();
+ const baselineMemoryAfter = this.getMemoryUsage();
const baselineTimeMs = baselineEnd - baselineStart;
const baselineAvgMs = baselineTimeMs / iterations;
+ const baselineMemoryUsed = Math.max(0, baselinePeakMemory - baselineMemoryBefore);
+
+ // Force garbage collection before Flash Attention benchmark
+ this.forceGC();
+
+ // Measure Flash Attention memory usage
+ const flashMemoryBefore = this.getMemoryUsage();
+ let flashPeakMemory = flashMemoryBefore;
+
+ // Benchmark Flash Attention
+ const flashStart = performance.now();
+ for (let i = 0; i < iterations; i++) {
+ this.flashAttention.computeRaw(query, keys, values);
+ // Sample memory periodically
+ if (i % 100 === 0) {
+ const currentMemory = this.getMemoryUsage();
+ if (currentMemory > flashPeakMemory) {
+ flashPeakMemory = currentMemory;
+ }
+ }
+ }
+ const flashEnd = performance.now();
+ const flashTimeMs = flashEnd - flashStart;
+ const flashAvgMs = flashTimeMs / iterations;
+ const flashMemoryUsed = Math.max(0, flashPeakMemory - flashMemoryBefore);
const speedup = baselineAvgMs / flashAvgMs;
const meetsTarget = speedup >= 2.49; // Minimum target: 2.49x
@@ -189,6 +220,13 @@ export class FlashAttentionOptimizer {
this.metrics.peakSpeedup = speedup;
}
+ // Update memory tracking metrics
+ this.metrics.totalBaselineMemory += baselineMemoryUsed;
+ this.metrics.totalOptimizedMemory += flashMemoryUsed;
+ if (flashPeakMemory > this.metrics.peakMemory) {
+ this.metrics.peakMemory = flashPeakMemory;
+ }
+
this.metrics.totalSpeedup += speedup;
if (meetsTarget) {
this.metrics.successfulOperations++;
@@ -198,12 +236,12 @@ export class FlashAttentionOptimizer {
flashAttention: {
averageTimeMs: flashAvgMs,
opsPerSecond: 1000 / flashAvgMs,
- memoryUsageBytes: undefined,
+ memoryUsageBytes: flashMemoryUsed,
},
baseline: {
averageTimeMs: baselineAvgMs,
opsPerSecond: 1000 / baselineAvgMs,
- memoryUsageBytes: undefined,
+ memoryUsageBytes: baselineMemoryUsed,
},
speedup,
meetsTarget,