Wire monitoring_service gRPC into web-gateway: proto compilation,
config/state/client plumbing, REST endpoint at /api/monitoring/live-metrics,
and a 3s polling bridge that broadcasts training_progress over WebSocket.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Adds a new CLI subcommand that connects to the monitoring service via
gRPC to display live training metrics. Supports single-snapshot mode
(--once), model filtering (--model), and configurable streaming
interval. Renders per-model epoch/loss table, GPU telemetry, hyperopt
progress, and health counters.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Prometheus client queries training/hyperopt/GPU/K8s metrics, gRPC service
exposes unary + server-streaming RPCs, 4 unit tests, zero clippy warnings.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Enable curiosity module by default (weight 0.0→0.1) to satisfy the
three-way gate (dueling + target + curiosity) that was blocking the
GPU experience collector CUDA kernel. This eliminates the 30-40% CPU
experience collection phase that was the main GPU idle bottleneck.
Additional changes:
- BatchSample API: train_step/compute_gradients now accept
Option<BatchSample> instead of Option<Vec<Experience>>, preserving
PER importance-sampling weights and indices through pre-sampling.
- Async PER pre-sampling: train_step_single_batch and
train_step_with_accumulation now pre-sample from the replay buffer
using a READ lock before acquiring the WRITE lock for GPU training.
This separates CPU sampling (~250μs) from GPU forward/backward (~3ms).
- Delete dead EpochPrefetcher: the binary (train_baseline_rl.rs) already
implements fold prefetching with background thread + mpsc channel +
GPU double-buffering, making the trainer's EpochPrefetcher redundant.
- Hyperopt DQN bounds: curiosity_weight min 0.0→0.01 so PSO can never
fully disable curiosity (which would re-gate the GPU collector).
10 files changed, -195 net lines. 2497 tests pass, 0 clippy warnings.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
New standalone monitoring_service that bridges Prometheus → gRPC,
enabling `fxt train monitor` and web-gateway WS streaming for
all training jobs (CI and service-managed).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The auto-detect heuristic used cpus/2 ("smart cap"), designed for
CPU-bound workloads. DQN/PPO trials are GPU-bound — each rayon
thread submits CUDA kernels and waits on cudaDeviceSynchronize(),
using minimal CPU. cpus-1 is the correct cap.
On L40S-1-48G (8 vCPU): 3 threads → 7 threads (2.3× more trials).
Also bumps CI CPU limit 7500m→8000m to expose all 8 cores.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The batch span processor needs a tokio runtime for gRPC transport and
periodic flush. Async services already have one via #[tokio::main], but
sync training binaries (hyperopt, train, evaluate) don't.
Previous approach (making binaries async with #[tokio::main]) caused
"Cannot start a runtime from within a runtime" panics because the ML
crate's internal code creates its own tokio runtimes for block_on().
New approach: build_otel_tracer() detects runtime context via
Handle::try_current(). If absent, it creates a dedicated 1-worker
multi-thread runtime stored in a process-lifetime OnceLock. The worker
thread actively polls the OTLP batch export task.
Reverts training binaries to sync fn main() so internal runtime creation
(hyperopt adapters, DQN/PPO trainers) continues working as before.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
All 6 training binaries (hyperopt_baseline_rl, hyperopt_baseline_supervised,
train_baseline_rl, train_baseline_supervised, evaluate_baseline,
evaluate_supervised) used sync fn main() but the OTLP batch exporter
requires a tokio runtime (tonic/hyper-util gRPC transport). This caused
an immediate panic on CI when OTEL_EXPORTER_OTLP_ENDPOINT was set.
Fix: #[tokio::main(flavor = "current_thread")] on all 6 binaries.
Also fix pre-existing clippy warnings (shadow, let_underscore_must_use,
doc_markdown, cognitive_complexity, integer_division, unsafe_code).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace CPU-side argmax (to_vec1 + iter enumerate max_by) with
GPU-native argmax(D). Single u32 scalar extraction instead of
full Q-value vector transfer.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace &Device::Cpu with &self.device for input and target tensor
creation in Mamba2 hyperopt adapter. Avoids unnecessary CPU→GPU
transfer during hyperparameter optimization.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Stream-aware ensemble that runs models on separate CUDA streams
for true GPU-level parallelism. Falls back to rayon on CPU.
Uses CudaStreamPool for synchronization.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
CUDA stream pool with CPU no-op fallback. Foundation for
StreamAwareEnsemble that runs models on separate CUDA streams.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Use predict_raw() to collect raw GPU tensors from adapters. Stack,
sigmoid, weighted-sum on GPU before single extraction. Falls back
to CPU path for adapters without tensor output.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Override predict_raw() in TGGN, TLOB, KAN, xLSTM, Diffusion adapters
to return raw GPU tensors. Enables GPU-side ensemble aggregation
instead of per-model CPU extraction.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace clamped.to_vec1() CPU loop with GPU-native floor/ceil/frac
operations. Removes 3 tensor transfers per forward pass (N floats
down + 3*N up). All index computation now stays on GPU.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Stack var_returns, var_residuals, mean_reward, var_reward into single
tensor before extraction. Uses broadcast_sub for GPU-native mean
centering instead of scalar round-trip.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Backward-compatible trait extension. Default predict_raw() wraps
predict() result with tensor: None. Adapters can override to return
raw GPU tensors for GPU-side ensemble aggregation.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The S3→PVC migration had several issues causing deploy failures:
- Stale binary-writer pod from previous failed deploys blocked new ones
(terminated pods can't be updated via kubectl apply)
- Services were applied BEFORE binaries written to PVC, so first deploy
or empty PVC caused pods to crash (binary not found)
- GPU overlay files in services/ dir were auto-applied by kubectl apply,
referencing nonexistent S3/minio secrets and PVCs
- Training job template mounted foxhunt-binaries PVC but training runs
on ci-training node — RWO PVC is bound to foxhunt node
Fixes:
- Reorder deploy: write binaries → apply manifests → rollout restart
- Clean up stale writer pods before creating new ones
- Move GPU overlays to gpu-overlays/ (manual apply only), update to PVC
- Add training-binaries PVC for GPU node, best-effort population
- Training job template uses initContainer to copy from training PVC
- set -e for critical path, set +e for optional training binary copy
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace per-param gradient norm extraction loop with batched
Tensor::stack pattern. Single GPU→CPU sync instead of one per
parameter tensor.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace per-param calculate_gradient_norm() loop with batched
Tensor::stack pattern. Single GPU→CPU sync instead of one per
parameter tensor.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace per-batch loss.to_vec0() in TFT training/validation loops
with GPU tensor accumulation. Single extraction per epoch + NaN guard
every 100 batches.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace per-batch .to_dtype(F64).to_scalar() in Liquid train/validate
with GPU tensor accumulation. Single extraction per epoch + NaN guard
every 100 batches. Removes ~100-500 GPU→CPU syncs per epoch.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace per-batch loss.to_scalar() in TLOB train_epoch/validate_epoch
with GPU tensor accumulation. Single extraction per epoch + NaN guard
every 100 batches.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The before_script was copying training binaries to ${CI_PROJECT_DIR}/bin/
which collides with the git checkout bin/fxt/ directory. chmod +x on the
fxt subdirectory fails with "Operation not permitted", crashing the job.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Runner #2 already has tags [kapsule, gpu] but CI jobs still used
kapsule-rl. Also removes stale minio-ca-cert volume from runner values.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Deploy stage needs compile-training artifacts to copy training binaries
to the foxhunt-binaries PVC alongside service binaries.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
MinIO pod was removed but compile-training still uploaded to it via rclone,
causing 403 Forbidden failures. Now:
- compile-training: saves binaries as CI artifacts (3 day TTL)
- deploy stage: copies training binaries to foxhunt-binaries PVC via binary-writer
- CI training jobs: use compile-training CI artifacts directly
- job-template.yaml: mounts foxhunt-binaries PVC (no initContainer)
- Removed MinIO deploy from deploy stage (no longer needed)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Both DQN and PPO eval paths used old 3-action index matching (0=Buy,
1=Sell, 2=Hold). Now uses FactoredAction.target_exposure() for
exposure-weighted returns and order-type-specific transaction costs.
PPO path had .to_int() which doesn't exist on FactoredAction.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
compute_reward_pnl() took raw action_idx (0=Buy,1=Sell,2=Hold) — wrong
with 45-action FactoredAction encoding. Now takes &FactoredAction and
uses order-type-specific transaction costs. Position tracking uses
action.exposure instead of old 3-way index match.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Hyperopt adapter and PPO benchmark still had num_actions: 3, which would
produce misconfigured PPO models when used with the 45-action FactoredAction
sampling path. Found by spec compliance review.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Use a purely linear network (no ReLU) for the completeness axiom test.
IG on linear functions is mathematically exact, so the test is
deterministic regardless of random weight initialization. Tighten
tolerance from 20% to 1% (f32 rounding only). Keep ReLU network in
the basic test for non-linear verification.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- trading_service: PPO predict() used TradingAction match but act() now
returns FactoredAction. Use target_exposure() mapped to 0-1 range.
- IG completeness axiom test: relax tolerance from 5% to 20% (random
weights with ReLU non-linearity and trapezoidal rule can exceed 5%).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Was using direction=idx/15 (3 groups of 15: Buy/Sell/Hold) — incompatible
with DQN's FactoredAction encoding. Now uses exposure_idx=idx/9 (5 groups
of 9: Short100/Short50/Flat/Long50/Long100) matching FactoredAction.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
sample_action(), act(), act_with_log_prob(), greedy_action() now return
FactoredAction instead of TradingAction. Fixes the architectural disconnect
where num_actions=45 output neurons were sampled through a 3-action bottleneck.
TrajectoryStep.action and TrajectoryBatch.actions now use FactoredAction.
Added FactoredAction::from_legacy() for backward compatibility in tests.
Updated all PPO consumers: trainers/ppo.rs, hyperopt/adapters/ppo.rs,
validation/ppo_adapter.rs, benchmark/ppo_benchmark.rs.
2487 tests pass, 0 clippy warnings.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Consolidates ExposureLevel, Urgency, FactoredAction from dqn/action_space.rs
and ppo/factored_action.rs into common/action.rs. Both dqn:: and ppo::
re-export for backward compatibility. Deletes ppo/factored_action.rs.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Changed from Status::unavailable to Status::unimplemented with path
forward (ml_training_service forwarding). Fixed supports_feature_importance
from true to false.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
calculate_feature_importance() returned hardcoded fabricated scores with
wrong feature names on every inference. Replaced with honest empty map.
Real importance is computed on-demand via integrated gradients.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Implements IntegratedGradients using Candle autograd (Var::from_tensor +
backward). Computes attributions by integrating input gradients along
interpolation path from baseline to input.
Verified via completeness axiom test: sum(attributions) ≈ F(x) - F(baseline).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The real_ prefix was misleading — there is no fake data loader.
Mechanical rename across 18 source files, no logic changes.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- DQN PER: defer td_errors to_vec1() after loss.to_scalar() — piggyback on
existing pipeline flush instead of forcing premature GPU→CPU stall
- PPO trajectories: capacity-hint Vec allocations, extend_flat_states methods,
states_flat field on TrajectoryBatch for zero-copy GPU upload
- TGGN validate(): batch N per-sample losses on GPU → single to_scalar() sync
(was N GPU→CPU syncs)
- Liquid backward(): batch grad-norm per-param sqr().sum_all() on GPU → single
to_scalar() sync (was N GPU→CPU syncs per optimizer step)
- Liquid validate(): same N→1 GPU sync reduction as TGGN
- DQN trainer: restore EpochPrefetcher/DoubleBufferedLoader API (wrongly deleted)
- train_baseline_rl: wire DoubleBuffer GPU pre-upload — after CPU prefetch
completes, immediately upload next fold to GPU via DqnGpuData::upload() so
next fold starts with data already resident on GPU
2478 tests pass, 0 clippy warnings, 0 compile errors.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
