937bbabb84
Batch upload: - Added indirect pad_states + indirect copy kernels to cubin - Batch source pointers uploaded as 8 x u64 via async HtoD to batch_ptr_buf - Kernels read source addresses from device indirection buffer - upload_batch_gpu replaced with upload_batch_ptrs → graph-captured indirect kernels - 8 ungraphed DtoD/kernel launches → 0 (captured in graph_forward) HER relabel (Random strategy): - Captured as graph_her (random_donors + inplace_relabel) - Uses stable addresses: donor_indices (pre-allocated), batch_ptr_buf[1] (indirect) - 2 ungraphed launches → 1 graph replay PER priority update: - Captured in graph_adam (after regime_scale) - Kernel changed to read indices/priorities from batch_ptr_buf[6..8] (indirect) - PER pointers uploaded via upload_per_ptrs before graph_adam replay - 1 ungraphed launch → 0 (captured in graph_adam) New CUDA kernels: - pad_states_indirect_kernel: reads src ptr from device buffer - indirect_copy_f32_kernel: f32 copy with indirect src - indirect_copy_i32_kernel: i32 copy with indirect src - per_update_priorities_kernel: changed to indirect indices/priorities Per-step operation count: 9 graph replays (forward, adam, ema, attention, iql, iqn, her) + 1 async HtoD (8 batch pointers, 64 bytes) + 1 async HtoD (tau, 4 bytes) + 1 async HtoD (adam_step, 4 bytes) + 1 async HtoD (iqn tau, 4 bytes) + 1 async HtoD (per ptrs, 16 bytes) = 9 graph replays + 5 async HtoD (92 bytes total) Zero ungraphed kernel launches in the per-step hot path. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Foxhunt
Production HFT trading system in Rust.
Architecture
The workspace contains 32 crates organized as follows:
Core Libraries (16)
| Crate | Purpose |
|---|---|
trading_engine |
Order processing, FIX 4.4, IB TWS, SIMD, RDTSC timing |
risk |
VaR, Kelly, circuit breakers, kill switches, compliance |
risk-data |
Risk data types and shared structures |
trading-data |
Trading data types |
ml |
DQN Rainbow, PPO, TFT, Mamba2, ensemble inference |
ml-data |
ML data types and feature definitions |
data |
Market data ingestion and storage |
backtesting |
Replay engine, strategy tester |
adaptive-strategy |
Ensemble execution, microstructure analysis |
common |
Shared types, resilience, error handling |
storage |
S3 and local model storage |
model_loader |
Model serialization and loading |
market-data |
Market data feed handlers |
database |
PostgreSQL access layer (SQLx) |
config |
Configuration management |
tli |
CLI commands and tooling |
Services (8)
| Service | Purpose |
|---|---|
backtesting_service |
gRPC backtesting service |
broker_gateway_service |
FIX routing, broker connectivity |
trading_service |
Core trading operations |
ml_training_service |
Model training orchestration |
data_acquisition_service |
Market data acquisition |
trading_agent_service |
Autonomous trading agents |
api_gateway |
gRPC API gateway with auth |
web-gateway |
Axum REST + WebSocket gateway |
Frontend
web-dashboard/ -- React 19 + TypeScript + Vite + TradingView charts.
Building
# Check compilation (no PostgreSQL required)
SQLX_OFFLINE=true cargo check --workspace
# Run tests for a specific crate
SQLX_OFFLINE=true cargo test -p <crate> --lib
# Clippy
SQLX_OFFLINE=true cargo clippy --workspace
ML Models
Four production model architectures on Candle v0.9.1 with CUDA:
- DQN Rainbow -- Deep Q-Network with prioritized replay, dueling heads, noisy nets
- PPO -- Proximal Policy Optimization with GAE and LSTM policies
- TFT -- Temporal Fusion Transformer for multi-horizon forecasting
- Mamba2 -- State space model for sequence prediction
Each model has a standalone trainer and a UnifiedTrainable adapter for the hyperopt pipeline.
Infrastructure
- Git: Gitea at
git.fxhnt.ai(Tailscale-only), Scaleway DEV1-S - Observability: OpenTelemetry OTLP (env
OTEL_EXPORTER_OTLP_ENDPOINT) - Database: PostgreSQL with SQLx offline mode for CI
License
Proprietary. All rights reserved.