fdb2a843b5
L40S nsys profile (commit0d630c799) identified three hot kernels: - backtest_state_gather: 37.2% / 60k calls - cuBLAS gemvx: 25.0% / 242k calls (val TLOB SGEMMs) - compute_expected_q: 12.9% / 1.4k calls Kernels #1 + cuBLAS #2 (folded together because most gemvx calls were val TLOB SGEMMs) — commit072ca4568: * Added backtest_state_gather_chunk kernel that writes [chunk_len, n_windows, padded_sd] directly into chunked_states_buf in a single launch with chunk_len*n_windows threads (1 thread per output row). * Refactored submit_dqn_step_loop_cublas Phase 1: per-step launch + DtoD copy loop replaced by a single batched call. Per-chunk launch count: 2*chunk_len -> 1 (1024x reduction at chunk_len=512). * Val TLOB now sized to DQN_BACKTEST_CHUNK_SIZE * n_windows so it runs once per chunk on chunked_states_buf at batch=chunk_len*n_windows instead of chunk_len separate calls. Per-chunk cuBLAS reduction: chunk_len*4=2048 -> 4 (~500x). * Single-step backtest_state_gather retained for evaluate() / evaluate_ppo() / evaluate_supervised() paths. Kernel #3 (compute_expected_q) — commit5f26c5751: * Replaced 3-pass softmax (max -> sum_exp -> normalize+expected+entropy) with online softmax (Page-Olshen running-max + running-sum). Inner atom loop: 3 -> 2 passes, ~33% fewer global loads of adv_a[z] per (action, branch) pair. * Bit-stable when atoms processed in fixed order. Underflow safety preserved analytically (no prob*log(prob) edge case). * ABI unchanged; no Rust caller migration needed. Deferred: Q/K/V SGEMM fusion (~50% of remaining ~500 val + ~300k training calls) gated by tlob_sdp_forward kernel ABI changes (stride parameter or unpacking step). ~200 LOC follow-up; recommend re-profiling post-deploy to confirm training-side TLOB still hot.
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.