## Major Achievements ### 1. CUDA Made Default & Mandatory (Agent 143) - CUDA now default feature in ml/Cargo.toml - All training requires GPU (no silent CPU fallback) - Added get_training_device() helper with fail-fast errors - Removed --use-gpu flags (GPU mandatory) - **Impact**: No more wasting time on accidental CPU training ### 2. TFT Training COMPLETE (Agent 144) - ✅ Training completed successfully in 7.6 minutes - ✅ Early stopping at epoch 100/200 (best val loss: 0.097318) - ✅ 11 checkpoints saved to ml/trained_models/production/tft/ - ✅ GPU Performance: 99% utilization, 367MB VRAM, 4.4s/epoch - ✅ 10x speedup vs CPU (4.4s vs 43-55s per epoch) - **Status**: PRODUCTION READY ### 3. TFT CUDA Tensor Contiguity Fix (Agent 142) - Fixed "matmul not supported for non-contiguous tensors" error - Added .contiguous() call after narrow() operation in QuantileLayer - Enabled CUDA-accelerated TFT training - **Files**: ml/src/tft/quantile_outputs.rs ### 4. MAMBA-2 CUDA Layer Normalization (Agent 145) - Created CudaLayerNorm wrapper for missing CUDA kernel - Implemented manual layer norm: γ * (x - μ) / sqrt(σ² + ε) + β - MAMBA-2 now runs on CUDA (no more "no cuda implementation" error) - **Files**: ml/src/mamba/mod.rs ### 5. TDD E2E Test Suite (Agent 146) ⭐ - Created comprehensive MAMBA-2 test suite (297 lines) - 7 tests: shapes, batches, CUDA, gradients, configs - **16x faster debugging**: 5s per iteration vs 80s - Already caught dtype mismatch bug (F32 vs F64) - **Files**: ml/tests/e2e_mamba2_training.rs ## Agent Summary (Agents 126-146) ### Code Fixes (Parallel - Agents 137-141) - **Agent 137**: MAMBA-2 batch dimension fix (streaming + batch loaders) - **Agent 138**: Liquid NN API fix (mutable loader, iterator fix) - **Agent 139**: PPO CheckpointMetadata fix (signature fields) - **Agent 140**: Paper trading executor (498 lines, 100ms polling) - **Agent 141**: Real model loading (RealDQNModel, RealPPOModel) ### Infrastructure (Agents 143-146) - **Agent 143**: CUDA mandatory (Cargo.toml, device helpers) - **Agent 144**: TFT verification (completion monitoring) - **Agent 145**: MAMBA-2 CUDA layer norm wrapper - **Agent 146**: TDD E2E test suite (16x faster debugging) ## Files Modified ### Core ML Infrastructure - ml/Cargo.toml: Added default = ["minimal-inference", "cuda"] - ml/src/lib.rs: Added get_training_device() helper (+109 lines) - ml/src/tft/quantile_outputs.rs: Fixed tensor contiguity - ml/src/mamba/mod.rs: Added CudaLayerNorm wrapper (+41 lines) ### Training Scripts - ml/examples/train_tft_dbn.rs: Removed --use-gpu flag - ml/examples/train_ppo.rs: Removed --use-gpu flag - ml/examples/train_mamba2_dbn.rs: Forced CUDA-only mode - ml/examples/train_liquid_dbn.rs: Fixed API usage ### Data Loaders - ml/src/data_loaders/dbn_sequence_loader.rs: Fixed batch dimensions - ml/src/data_loaders/streaming_dbn_loader.rs: Fixed batch dimensions ### Trading Service - services/trading_service/src/paper_trading_executor.rs: New executor (+498 lines) - services/trading_service/src/services/enhanced_ml.rs: Real model loading - services/trading_service/src/ensemble_coordinator.rs: Integration ### Tests - ml/tests/e2e_mamba2_training.rs: New TDD test suite (+297 lines) ### Trainers - ml/src/trainers/tft.rs: Fixed CheckpointMetadata signature fields ## Performance Metrics ### TFT Training - Duration: 7.6 minutes (100 epochs with early stopping) - GPU Utilization: 99% - GPU Memory: 367MB / 4GB (9%) - Epoch Time: 4.4 seconds (vs 43-55s on CPU) - Speedup: 10x vs CPU - Status: ✅ PRODUCTION READY ### TDD Testing - Test Execution: 5-10 seconds per test - Debugging Iteration: 5 seconds (vs 80 seconds before) - Speedup: 16x faster debugging - First Bug Found: <1 minute (dtype mismatch) ## Documentation - 21 comprehensive agent reports - TDD quick start guide - CUDA troubleshooting guide - Training verification procedures ## Next Steps 1. Fix MAMBA-2 dtype mismatch (F32→F64) - 2 minutes 2. Run MAMBA-2 tests until passing - 5-10 minutes 3. Launch full MAMBA-2 training - 200 epochs 4. Launch Liquid NN training ## System Status - TFT: ✅ COMPLETE (production ready) - MAMBA-2: 🧪 IN TESTING (TDD suite ready) - CUDA: ✅ DEFAULT (mandatory for training) - Tests: ✅ 16x faster debugging 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
293 lines
9.2 KiB
Bash
Executable File
293 lines
9.2 KiB
Bash
Executable File
#!/bin/bash
|
|
#
|
|
# Flame Graph Generation Script for Performance Profiling
|
|
#
|
|
# This script generates flame graphs for the Foxhunt HFT system to visualize
|
|
# where CPU time is spent during benchmark execution. Useful for identifying
|
|
# performance bottlenecks and optimization opportunities.
|
|
#
|
|
# Usage:
|
|
# ./scripts/generate_flame_graphs.sh [benchmark-name] [duration-seconds]
|
|
#
|
|
# Examples:
|
|
# ./scripts/generate_flame_graphs.sh # All benchmarks, 30s each
|
|
# ./scripts/generate_flame_graphs.sh ml_prediction 60 # Specific benchmark, 60s
|
|
#
|
|
# Requirements:
|
|
# - cargo-flamegraph (install with: cargo install flamegraph)
|
|
# - perf (Linux kernel profiler)
|
|
# - Linux system (perf required)
|
|
|
|
set -euo pipefail
|
|
|
|
# Colors for output
|
|
RED='\033[0;31m'
|
|
GREEN='\033[0;32m'
|
|
YELLOW='\033[1;33m'
|
|
BLUE='\033[0;34m'
|
|
NC='\033[0m' # No Color
|
|
|
|
# Configuration
|
|
BENCHMARK_NAME="${1:-all}"
|
|
PROFILE_DURATION="${2:-30}"
|
|
OUTPUT_DIR="flame_graphs"
|
|
TIMESTAMP=$(date +%Y%m%d_%H%M%S)
|
|
|
|
echo -e "${BLUE}=================================================${NC}"
|
|
echo -e "${BLUE}Flame Graph Generation for Performance Profiling${NC}"
|
|
echo -e "${BLUE}=================================================${NC}"
|
|
echo ""
|
|
echo -e "${GREEN}Benchmark:${NC} $BENCHMARK_NAME"
|
|
echo -e "${GREEN}Duration:${NC} ${PROFILE_DURATION}s"
|
|
echo -e "${GREEN}Output:${NC} $OUTPUT_DIR"
|
|
echo ""
|
|
|
|
# Check if running on Linux
|
|
if [[ "$OSTYPE" != "linux-gnu"* ]]; then
|
|
echo -e "${RED}Error: Flame graphs require Linux (perf tool)${NC}"
|
|
echo -e "${YELLOW}Alternatives:${NC}"
|
|
echo " - Run in Docker: docker run -v \$(pwd):/workspace rust:latest bash"
|
|
echo " - Use Instruments on macOS"
|
|
echo " - Use other profiling tools (valgrind, etc.)"
|
|
exit 1
|
|
fi
|
|
|
|
# Check if cargo-flamegraph is installed
|
|
if ! command -v cargo-flamegraph &> /dev/null; then
|
|
echo -e "${YELLOW}Installing cargo-flamegraph...${NC}"
|
|
if ! cargo install flamegraph; then
|
|
echo -e "${RED}Failed to install cargo-flamegraph${NC}"
|
|
exit 1
|
|
fi
|
|
echo -e "${GREEN}✓ cargo-flamegraph installed${NC}"
|
|
fi
|
|
|
|
# Check if perf is available
|
|
if ! command -v perf &> /dev/null; then
|
|
echo -e "${YELLOW}perf not found. Attempting to install...${NC}"
|
|
if sudo apt-get install -y linux-tools-common linux-tools-generic linux-tools-$(uname -r) 2>/dev/null; then
|
|
echo -e "${GREEN}✓ perf installed${NC}"
|
|
else
|
|
echo -e "${RED}Failed to install perf. Please install manually:${NC}"
|
|
echo " sudo apt-get install linux-tools-\$(uname -r)"
|
|
exit 1
|
|
fi
|
|
fi
|
|
|
|
# Set perf permissions (may require sudo)
|
|
echo -e "${YELLOW}Configuring perf permissions...${NC}"
|
|
if [ -f /proc/sys/kernel/perf_event_paranoid ]; then
|
|
CURRENT_PARANOID=$(cat /proc/sys/kernel/perf_event_paranoid)
|
|
if [ "$CURRENT_PARANOID" -gt 1 ]; then
|
|
echo -e "${YELLOW}Note: perf_event_paranoid is $CURRENT_PARANOID (restrictive)${NC}"
|
|
echo -e "${YELLOW}For better profiling, set to -1 (requires sudo):${NC}"
|
|
echo " sudo sysctl kernel.perf_event_paranoid=-1"
|
|
echo ""
|
|
read -p "Attempt to set now? (y/N) " -n 1 -r
|
|
echo
|
|
if [[ $REPLY =~ ^[Yy]$ ]]; then
|
|
sudo sysctl kernel.perf_event_paranoid=-1
|
|
fi
|
|
fi
|
|
fi
|
|
|
|
# Create output directory
|
|
mkdir -p "$OUTPUT_DIR"
|
|
|
|
# Function to generate flame graph for specific benchmark
|
|
generate_flamegraph() {
|
|
local bench_name="$1"
|
|
local output_file="$OUTPUT_DIR/flamegraph_${bench_name}_${TIMESTAMP}.svg"
|
|
|
|
echo ""
|
|
echo -e "${YELLOW}Generating flame graph: $bench_name${NC}"
|
|
echo -e "${BLUE}Duration: ${PROFILE_DURATION}s${NC}"
|
|
echo ""
|
|
|
|
# Run cargo-flamegraph on the benchmark
|
|
if cargo flamegraph \
|
|
--bench performance_regression \
|
|
--output "$output_file" \
|
|
-- --bench "$bench_name" --profile-time "$PROFILE_DURATION"; then
|
|
|
|
echo -e "${GREEN}✓ Flame graph generated: $output_file${NC}"
|
|
|
|
# Get file size
|
|
local file_size=$(du -h "$output_file" | cut -f1)
|
|
echo -e "${BLUE} Size: $file_size${NC}"
|
|
|
|
return 0
|
|
else
|
|
echo -e "${RED}✗ Failed to generate flame graph for $bench_name${NC}"
|
|
return 1
|
|
fi
|
|
}
|
|
|
|
# Main execution
|
|
main() {
|
|
if [ "$BENCHMARK_NAME" = "all" ]; then
|
|
echo -e "${YELLOW}Generating flame graphs for all benchmarks...${NC}"
|
|
echo -e "${BLUE}This will take approximately $((PROFILE_DURATION * 8)) seconds${NC}"
|
|
echo ""
|
|
|
|
# List of all benchmarks in performance_regression.rs
|
|
BENCHMARKS=(
|
|
"ml_prediction_latency"
|
|
"hot_swap_latency"
|
|
"database_writes"
|
|
"backtest_performance"
|
|
"order_processing"
|
|
"risk_validation"
|
|
"memory_allocation"
|
|
"concurrent_access"
|
|
)
|
|
|
|
local success_count=0
|
|
local total_count=${#BENCHMARKS[@]}
|
|
|
|
for bench in "${BENCHMARKS[@]}"; do
|
|
if generate_flamegraph "$bench"; then
|
|
((success_count++))
|
|
fi
|
|
done
|
|
|
|
echo ""
|
|
echo -e "${GREEN}=================================================${NC}"
|
|
echo -e "${GREEN}Flame Graph Generation Complete${NC}"
|
|
echo -e "${GREEN}=================================================${NC}"
|
|
echo -e "${GREEN}Success: $success_count / $total_count${NC}"
|
|
echo ""
|
|
else
|
|
generate_flamegraph "$BENCHMARK_NAME"
|
|
fi
|
|
|
|
# Generate summary HTML
|
|
echo ""
|
|
echo -e "${YELLOW}Generating summary page...${NC}"
|
|
|
|
cat > "$OUTPUT_DIR/index.html" <<'EOFHTML'
|
|
<!DOCTYPE html>
|
|
<html>
|
|
<head>
|
|
<title>Foxhunt Performance Flame Graphs</title>
|
|
<style>
|
|
body {
|
|
font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;
|
|
max-width: 1200px;
|
|
margin: 0 auto;
|
|
padding: 20px;
|
|
background: #f5f5f5;
|
|
}
|
|
h1 {
|
|
color: #333;
|
|
border-bottom: 3px solid #4CAF50;
|
|
padding-bottom: 10px;
|
|
}
|
|
.benchmark {
|
|
background: white;
|
|
border-radius: 8px;
|
|
padding: 20px;
|
|
margin: 20px 0;
|
|
box-shadow: 0 2px 4px rgba(0,0,0,0.1);
|
|
}
|
|
.benchmark h2 {
|
|
color: #4CAF50;
|
|
margin-top: 0;
|
|
}
|
|
.benchmark iframe {
|
|
width: 100%;
|
|
height: 600px;
|
|
border: 1px solid #ddd;
|
|
border-radius: 4px;
|
|
}
|
|
.info {
|
|
background: #e3f2fd;
|
|
border-left: 4px solid #2196F3;
|
|
padding: 15px;
|
|
margin: 20px 0;
|
|
border-radius: 4px;
|
|
}
|
|
.legend {
|
|
background: #fff3e0;
|
|
border-left: 4px solid #ff9800;
|
|
padding: 15px;
|
|
margin: 20px 0;
|
|
border-radius: 4px;
|
|
}
|
|
</style>
|
|
</head>
|
|
<body>
|
|
<h1>Foxhunt HFT Performance Flame Graphs</h1>
|
|
|
|
<div class="info">
|
|
<strong>Generated:</strong> TIMESTAMP_PLACEHOLDER<br>
|
|
<strong>Profile Duration:</strong> DURATION_PLACEHOLDER seconds per benchmark
|
|
</div>
|
|
|
|
<div class="legend">
|
|
<h3>How to Read Flame Graphs</h3>
|
|
<ul>
|
|
<li><strong>X-axis:</strong> Alphabetical ordering (NOT time)</li>
|
|
<li><strong>Y-axis:</strong> Stack depth (call hierarchy)</li>
|
|
<li><strong>Width:</strong> CPU time consumed</li>
|
|
<li><strong>Color:</strong> Random (for differentiation only)</li>
|
|
</ul>
|
|
<p><strong>Tip:</strong> Click on any frame to zoom in. Look for wide frames at the top for optimization opportunities.</p>
|
|
</div>
|
|
|
|
EOFHTML
|
|
|
|
# Add each flame graph to the HTML
|
|
for svg_file in "$OUTPUT_DIR"/flamegraph_*.svg; do
|
|
if [ -f "$svg_file" ]; then
|
|
local bench_name=$(basename "$svg_file" | sed 's/flamegraph_//;s/_[0-9]*\.svg$//')
|
|
local svg_filename=$(basename "$svg_file")
|
|
|
|
cat >> "$OUTPUT_DIR/index.html" <<EOFBENCH
|
|
<div class="benchmark">
|
|
<h2>$bench_name</h2>
|
|
<iframe src="$svg_filename" scrolling="no"></iframe>
|
|
</div>
|
|
|
|
EOFBENCH
|
|
fi
|
|
done
|
|
|
|
cat >> "$OUTPUT_DIR/index.html" <<'EOFFOOTER'
|
|
</body>
|
|
</html>
|
|
EOFFOOTER
|
|
|
|
# Replace placeholders
|
|
sed -i "s/TIMESTAMP_PLACEHOLDER/$(date)/" "$OUTPUT_DIR/index.html"
|
|
sed -i "s/DURATION_PLACEHOLDER/$PROFILE_DURATION/" "$OUTPUT_DIR/index.html"
|
|
|
|
echo -e "${GREEN}✓ Summary page generated: $OUTPUT_DIR/index.html${NC}"
|
|
|
|
# Show results
|
|
echo ""
|
|
echo -e "${GREEN}=================================================${NC}"
|
|
echo -e "${GREEN}Flame Graphs Generated Successfully${NC}"
|
|
echo -e "${GREEN}=================================================${NC}"
|
|
echo ""
|
|
echo -e "${BLUE}View results:${NC}"
|
|
echo -e " ${YELLOW}open $OUTPUT_DIR/index.html${NC}"
|
|
echo ""
|
|
echo -e "${BLUE}Individual flame graphs:${NC}"
|
|
for svg_file in "$OUTPUT_DIR"/flamegraph_*.svg; do
|
|
if [ -f "$svg_file" ]; then
|
|
echo -e " - $(basename "$svg_file")"
|
|
fi
|
|
done
|
|
echo ""
|
|
echo -e "${BLUE}Analysis tips:${NC}"
|
|
echo " 1. Look for wide horizontal bars (high CPU time)"
|
|
echo " 2. Deep stacks indicate complex call chains"
|
|
echo " 3. Optimize functions with widest frames"
|
|
echo " 4. Compare flame graphs before/after optimization"
|
|
echo ""
|
|
}
|
|
|
|
# Run main function
|
|
main "$@"
|