gitadmin/foxhunt
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
3db41edf704bd1663309d2bb4dd3eee263e14e77
foxhunt/GPU_MEMORY_PROFILE_REPORT.md
jgrusewski 35feadf55e 🚀 Wave 160 Phase 6: CUDA Mandatory + TDD Testing + TFT Complete (21 Agents) 
## 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>
2025-10-14 23:13:34 +02:00

4.9 KiB
Raw Blame History

GPU Memory Profile Report - RTX 3050 Ti (4GB VRAM)

Generated: 2025-10-14 19:38:02 UTC GPU: NVIDIA GeForce RTX 3050 Ti Laptop VRAM: 4096 MB total, 3669 MB free at start

Executive Summary

This report profiles GPU VRAM usage for all ML models using direct nvidia-smi measurements.

  • DQN: 135.0 MB peak VRAM, batch size 64 (training), batch size 128 (inference) - ✅ Safe
  • PPO: 135.0 MB peak VRAM, batch size 64 (training), batch size 128 (inference) - ✅ Safe
  • MAMBA-2: 167.0 MB peak VRAM, batch size 32 (training), batch size 64 (inference) - ✅ Safe
  • TFT: 167.0 MB peak VRAM, batch size 8 (training), batch size 16 (inference) - ✅ Safe
  • Liquid NN: 167.0 MB peak VRAM, batch size 64 (training), batch size 128 (inference) - ✅ Safe

Detailed Model Profiles

DQN

  • Parameters: 83717
  • Base VRAM: 103.0 MB
  • Peak VRAM: 135.0 MB
  • Status: ✅ Safe
  • Max Safe Batch Size: 512
  • Training Batch Size: 64
  • Inference Batch Size: 128

Batch Size Tests

Batch Size VRAM (MB) Status
1 135.0 (3%) ✅ Success
8 135.0 (3%) ✅ Success
16 135.0 (3%) ✅ Success
32 135.0 (3%) ✅ Success
64 135.0 (3%) ✅ Success
128 135.0 (3%) ✅ Success
256 135.0 (3%) ✅ Success
512 135.0 (3%) ✅ Success

PPO

  • Parameters: 165376
  • Base VRAM: 135.0 MB
  • Peak VRAM: 135.0 MB
  • Status: ✅ Safe
  • Max Safe Batch Size: 256
  • Training Batch Size: 64
  • Inference Batch Size: 128

Batch Size Tests

Batch Size VRAM (MB) Status
1 135.0 (3%) ✅ Success
8 135.0 (3%) ✅ Success
16 135.0 (3%) ✅ Success
32 135.0 (3%) ✅ Success
64 135.0 (3%) ✅ Success
128 135.0 (3%) ✅ Success
256 135.0 (3%) ✅ Success

MAMBA-2

  • Parameters: 786432
  • Base VRAM: 135.0 MB
  • Peak VRAM: 167.0 MB
  • Status: ✅ Safe
  • Max Safe Batch Size: 64
  • Training Batch Size: 32
  • Inference Batch Size: 64

Batch Size Tests

Batch Size VRAM (MB) Status
1 135.0 (3%) ✅ Success
4 135.0 (3%) ✅ Success
8 135.0 (3%) ✅ Success
16 135.0 (3%) ✅ Success
32 135.0 (3%) ✅ Success
64 167.0 (4%) ✅ Success

TFT

  • Parameters: 6291456
  • Base VRAM: 167.0 MB
  • Peak VRAM: 167.0 MB
  • Status: ✅ Safe
  • Max Safe Batch Size: 32
  • Training Batch Size: 8
  • Inference Batch Size: 16

Batch Size Tests

Batch Size VRAM (MB) Status
1 167.0 (4%) ✅ Success
2 167.0 (4%) ✅ Success
4 167.0 (4%) ✅ Success
8 167.0 (4%) ✅ Success
16 167.0 (4%) ✅ Success
32 167.0 (4%) ✅ Success

Liquid NN

  • Parameters: 83456
  • Base VRAM: 167.0 MB
  • Peak VRAM: 167.0 MB
  • Status: ✅ Safe
  • Max Safe Batch Size: 256
  • Training Batch Size: 64
  • Inference Batch Size: 128

Batch Size Tests

Batch Size VRAM (MB) Status
1 167.0 (4%) ✅ Success
8 167.0 (4%) ✅ Success
16 167.0 (4%) ✅ Success
32 167.0 (4%) ✅ Success
64 167.0 (4%) ✅ Success
128 167.0 (4%) ✅ Success
256 167.0 (4%) ✅ Success

Memory Budget Allocation

Training (Single Model)

Model Peak VRAM Training Batch Status
DQN 135.0 MB 64 ✅
PPO 135.0 MB 64 ✅
MAMBA-2 167.0 MB 32 ✅
TFT 167.0 MB 8 ✅
Liquid NN 167.0 MB 64 ✅

Inference (Multi-Model Ensemble)

  • Total VRAM for all models: 707.0 MB
  • Available VRAM: 4096.0 MB
  • Can load all models: ✅ Yes

Recommendations

Training

  1. Train one model at a time - Use recommended batch sizes above
  2. Monitor VRAM - Run watch -n1 nvidia-smi during training
  3. Use gradient accumulation for TFT model (small batch size)
  4. Enable mixed precision (FP16) to reduce VRAM by ~40%
  5. Clear CUDA cache between model switches: torch.cuda.empty_cache()

Inference

  1. All models can be loaded simultaneously for ensemble inference
  2. Use batch inference with recommended batch sizes

Expected vs Actual VRAM Usage

Model Expected Range (MB) Actual (MB) Status
DQN 50-150 135.0 ✅ Within range
PPO 50-200 135.0 ✅ Within range
MAMBA-2 150-500 167.0 ✅ Within range
TFT 1500-2500 167.0 ⚠️ Lower
Liquid NN 100-300 167.0 ✅ Within range

Agent: 133 (GPU Memory Profiling) Command: cargo run -p ml --example gpu_memory_benchmark --release --features cuda

Reference in New Issue View Git Blame Copy Permalink