✅ Tasks
This page tracks project work that may or may not be connected to a specific experiment. Tasks can describe code, documentation, operations, slides, meetings, server/device setup, or research planning.
The task log is generated from the shared task registry in site/src/data/tasks.js. The same registry now drives the project roadmap on the overview page.
A task with status todo is unassigned. Once someone starts working on it, set status: 'inprogress', add that person to owners, and fill start_date. Once finished, set status: 'completed' and fill end_date. due_date may be left empty, which renders as UNDEFINED.
Each task is classified by three axes:
stage: lifecycle position, such as research, implementation, validation, integration, deployment, or publication.target: the artifact or system being changed, such asenabol,software,hls4ml,cpp_backend,csim,fpga, orpaper.action: the kind of work, such as design, development, documentation, experiment, testing, integration, analysis, or writing.
Active Tasks
ENB-013In progressCreate reusable precision-map records
Promote repeated fixed-point formats into stable PREC-* records once multiple experiments reuse the same precision maps.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- UNDEFINED
HLS4ML-008In progressHLS4ML - Implement synthesizable controller kernels
CTRL-NONE, CTRL-GT-ORDER-0, and CTRL-GT-ORDER-1 are implemented and wired. GT-0/GT-1 now use raw optimizer update geometry plus division-free candidate search over α²·η²·||ΔG||² ≤ χ²·(||Δθ_raw||²+ε²), with a nonzero alpha candidate floor. CTRL-GT-ORDER-2/2-QA remain.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- UNDEFINED
HLS4ML-009In progressHLS4ML - Add global metric reductions and alpha synchronization
Raw-update controller sensing is separated from controller law (raw_update_sensor_order0 → global_throttle_*_law). The writer emission now runs backpass → SGD proposal → raw-update/dgrad reduction → law → alpha-scaled apply. Per-layer raw update norms and ΔG norms are summed into global accumulators before the law call. CTRL-NONE skips curvature sensing.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-022In progressHLS4ML - Implement CTRL-GT-ORDER-1 controller
Implemented global_throttle_order1_law with persistent alpha_state. Uses the same raw-update controller geometry and writer emission as GT-0. KAlpha config field added to bridge and config struct. Pending CSIM validation after the raw-update geometry fix.
- Due
- UNDEFINED
- Start
- 2026-05-23
- End
- UNDEFINED
ENB-010In progressExpand software stress tests beyond the first sweep
Run distribution shift, transient spike, rail saturation, and update dead-zone tests across enough benchmarks to stress the controller.
- Due
- UNDEFINED
- Start
- 2026-05-20
- End
- UNDEFINED
HLS4ML-037In progressHLS4ML - Add optional trainable CSIM trace logging
Current slice implemented: generated trainable CSIM writes metadata-bearing comma-delimited .dat traces under tb_data/training/ for loss, alpha, and per-layer epoch-level parameter traces such as dense0/weights.dat and dense0/biases.dat. Repeated epoch/sample/global_step/sample_index columns allow sparse traces with different cadences. Broader trace selection for gradients and controller internals remains open.
- Due
- UNDEFINED
- Start
- 2026-05-23
- End
- UNDEFINED
ENB-014In progressValidate Conv2D backprop path in hls4ml trainable backend
Re-run MSWC-style tests with trainable convolutional layers, checking gradient flow through frozen dense layers and correctness of conv2d backpass.
- Due
- 2026-05-25
- Start
- 2026-05-18
- End
- UNDEFINED
ENB-015In progressImplement and validate BatchNorm backprop
Correct BN backprop even when BN parameters are frozen, because gradients must propagate through BN into earlier trainable layers.
- Due
- 2026-05-25
- Start
- 2026-05-18
- End
- UNDEFINED
ENB-016In progressFork hls4ml:main, create :trainable or :enabol branch and set up repo
Fork the hls4ml repo and set up the branch structure and CI paths for integrating the trainable controller code path into hls4ml templates and vivado writer.
- Due
- 2026-05-28
- Start
- 2026-05-21
- End
- 2026-05-21
ENB-016bIn progressAnalyze the hls4ml repo and trace an updated strategy for integrating our controller.
Port the selected global throttle path from software experiments into the hls4ml trainable code path and code generation templates.
- Due
- 2026-05-28
- Start
- 2026-05-21
- End
- UNDEFINED
ENB-020In progressMaintain project meeting notes and task status
Keep the project memory current as the work shifts from software ablations to hls4ml and hardware.
- Due
- UNDEFINED
- Start
- 2026-05-20
- End
- UNDEFINED
HLS4ML-025UnassignedHLS4ML - Reintroduce kappa safety budgets as optional bounds
Treat kappa as an optional safety bound or gain-rail budget, not as the primary learning throttle. Decide global versus per-layer kappa policy before adding hardware code.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-010UnassignedHLS4ML - Evaluate delayed-alpha pipelined update schedule
Analyze the one-step-delayed alpha schedule where alpha_t is computed from the current backpass but alpha_{t-1} is used to apply stored updates during the next forward traversal.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-023UnassignedHLS4ML - Implement CTRL-GT-ORDER-2 controller
Add second-order controller state after GT1. Validate alpha/velocity state update and fixed-point rails against software traces.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-024UnassignedHLS4ML - Implement CTRL-GT-ORDER-2-QA controller
Add the quantization-aware lower alpha bound once precision inference is available. This should use explicit update quantum metadata rather than ad hoc bit inspection.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-027UnassignedHLS4ML - Add binary crossentropy loss endpoints
Add binary_crossentropy and binary_crossentropy_from_logits metadata/templates, including final-sigmoid skip semantics and fixed-point/PWL helper review.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-028UnassignedHLS4ML - Add categorical crossentropy loss endpoints
Add categorical_crossentropy and categorical_crossentropy_from_logits, including final-softmax skip semantics and stable gradient seed policy.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-030UnassignedHLS4ML - Add activation backpasses
Port/rewrite activation backpasses for ReLU, sigmoid, tanh, softmax where appropriate. Start with the activation needed by the two-Dense validation model.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-036UnassignedHLS4ML - Add optimizer-state allocation framework
Generalize optimizer state allocation so SGD, Adam, and future optimizers do not hide static arrays inside layer kernels.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-026UnassignedHLS4ML - Add MAE loss endpoint support
Port or rewrite MAE loss and gradient seed after half_mse is validated. Define zero-difference subgradient policy explicitly.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-029UnassignedHLS4ML - Add secondary regression/divergence losses
Future support for MAPE, MSLE, cosine similarity, and KL divergence. These should not block the first Dense trainable compile path.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-031UnassignedHLS4ML - Add BatchNorm backpass
Add BatchNorm backward support after Dense and activation paths are stable. Reconcile old BatchNorm backprop code with current hls4ml BatchNormalization layer metadata.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-032UnassignedHLS4ML - Add Conv1D and Conv2D backpasses
Add convolution backpasses only after the Dense path proves the graph-flow/template interface. Start from Conv1D or the smallest Conv2D fixture, not from the full deprecated generic code.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-033UnassignedHLS4ML - Add pooling backpasses
Future support for max/average/global pooling gradients. This likely needs cached argmax or mask policy for max pooling.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-034UnassignedHLS4ML - Add layout-only backpasses
Port/rewrite simple layout gradients such as Flatten, Reshape, ZeroPadding, Cropping, and UpSampling after activation traversal is working.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-035UnassignedHLS4ML - Add Adam optimizer support
Future Adam support with explicit state buffers and bias-correction policy. The deprecated code mixes double/std::pow paths and large static arrays, so this must be rewritten carefully before synthesis.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-013UnassignedHLS4ML - Validate one-layer Dense trainable CSIM
Run the minimal one-dense affine regression case through generated trainable firmware in CSIM and compare forward values, loss, gradients, alpha, and updated weights against ENABOL software traces.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
ENB-017UnassignedRun CSIM validation with global throttle (CTRL-GT-ORDER-0, division-free fix)
GT-0 CSIM failed: two ap_fixed divisions (sqrt + divide) crashed CSIM at step 1. Fixed by replacing α = χ/(ηC+ε) with division-free inequality search α²·η²·||ΔG||² ≤ χ²·(||Δθ_raw||² + ε²). The raw update is the optimizer proposal before alpha/quantized assignment; for SGD it already includes η. Candidate table of binary-fraction α values is searched in descending order with a nonzero alpha floor. Validate: compile 1-Dense model with Kind: CTRL-GT-ORDER-0, run CSIM, confirm α selection, no CSIM errors, loss drops.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-014UnassignedHLS4ML - Extend trainable CSIM to two Dense layers with activation
After the one-layer Dense path is correct, validate reverse traversal through Dense -> activation -> Dense before moving to convolution and BatchNorm paths.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
ENB-023UnassignedInfer hls4ml trainable precisions from ENABOL model, dataset, optimizer, and controller
Replace placeholder/default trainable precision fields with an inference pass that uses model ranges, dataset ranges, learning rate, batch size, optimizer state, controller metrics, loss scale, and fixed-point safety margins.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
ENB-019UnassignedRun FPGA hardware drift-recovery experiment
Collect the real hardware baseline, post-drift initial point, and post-training recovery point for controller-enabled learning.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
ENB-018UnassignedRun HLS synthesis and resource validation
Check resource impact, timing, and generated design viability after CSIM validates behavior.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
ENB-021UnassignedPrepare paper narrative, plots, and handoff package
Turn the final validated story into paper-ready figures, narrative, and handoff documentation.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
ENB-001CompletedBuild the ENABOL documentation and project registry system
Created the Docusaurus structure, metadata components, people registry, task board, meeting calendar, and implementation/experiment hubs.
- Due
- 2026-05-19
- Start
- 2026-05-1
- End
- 2026-05-16
ENB-003CompletedStudy global throttle
Separated static κ-style representational rails from dynamic trajectory control and documented why global throttling preserves update direction.
- Due
- 2026-05-20
- Start
- 2026-05-09
- End
- 2026-05-18
ENB-002CompletedStudy κ-budgeting failure mode
Separated static κ-style representational rails from dynamic trajectory control and documented why global throttling preserves update direction.
- Due
- 2026-05-09
- Start
- 2026-05-01
- End
- 2026-05-09
ENB-028DoneDocument trainable precision schema relationships
Added the first precision relationship document. Unknown trainable precision dependencies are explicitly marked with ?, and the known rule for controller_metric records that dtheta_sq needs fractional bits for squared parameter movement: F_controller_metric >= 2*F_theta + guard bits.
- Due
- UNDEFINED
- Start
- 2026-05-23
- End
- 2026-05-23
HLS4ML-001DoneHLS4ML - Define the trainable HLSConfig schema
Done in hls4ml-trainable: HLSConfig now has normalized Model.Training defaults and accessors for trainable mode, loss, optimizer, controller, trainable precision, recognized precision fields, and per-layer Training overrides.
- Due
- UNDEFINED
- Start
- 2026-05-21
- End
- 2026-05-22
ENB-004CompletedDefine the controller registry and stability stress-test catalog
Established CTRL-* pages and ST-* stress-test pages as stable IDs for experiments and task tracking.
- Due
- 2026-05-20
- Start
- 2026-05-09
- End
- 2026-05-18
HLS4ML-039DoneHLS4ML - Use raw-update geometry for global throttle control
Replaced actual/throttled parameter-delta controller geometry with raw optimizer update geometry. SGD now runs before the law, raw_update_sensor_order0 sums ||Δθ_raw||² and ||ΔG||² globally, GT-0/GT-1 keep the division-free squared inequality, alpha uses a nonzero candidate floor, and controller.dat logs raw/controlled/actual update norms plus feasibility telemetry.
- Due
- UNDEFINED
- Start
- 2026-05-23
- End
- 2026-05-23
HLS4ML-015DoneHLS4ML - Port minimal trainable static headers
Done: minimal static support files now live under templates/vivado/trainable/{common,losses,backprop,optimizers,controllers}. This includes trainable_trace.h, mse.h, nnet_dense_backprop.h, sgd.h, and global_throttle.h.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-22
HLS4ML-016DoneHLS4ML - Implement half-MSE loss template and dLdy seed
Done: mse.h implements mse and half_mse using one shift-based core, with half_mse seeding dL/dy = y_hat - y and mse seeding dL/dy = 2 * (y_hat - y).
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-22
HLS4ML-017DoneHLS4ML - Emit trainable top-level IO for loss and training controls
Done for the minimal path: generated top-level firmware receives ground truth, scalar loss output, alpha output, train_enable, reset_accumulators, and batch_end. The testbench trains from tb_output_predictions.dat, while the Python bridge supplies zero trainable tensors with train_enable=false for inference-safe calls.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-22
HLS4ML-018DoneHLS4ML - Implement minimal Dense backpass without controller
Done: the first Dense backward kernel computes dL/dx, dL/dW, and dL/db accumulation and the writer emits the corresponding call for the resolved backward traversal. Global throttle, kappa, Adam, and delayed-alpha scheduling remain future work.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-22
HLS4ML-019DoneHLS4ML - Implement raw SGD update proposal for Dense
Done: sgd.h computes raw weight/bias update proposals from averaged gradients and learning rate, and writer emission calls SGD before CTRL-NONE alpha application.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-22
HLS4ML-021DoneHLS4ML - Implement CTRL-GT-ORDER-0 controller
Implemented CTRL-GT-ORDER-0 as a global law fed by raw_update_sensor_order0. The writer sums per-layer ||Δθ_raw||² and ||ΔG||² contributions into global accumulators, then GT-0 selects the largest candidate alpha satisfying α²·η²·||ΔG||² ≤ χ²·(||Δθ_raw||²+ε²). This avoids sqrt/division and avoids the self-locking actual-update denominator.
- Due
- UNDEFINED
- Start
- 2026-05-23
- End
- 2026-05-23
HLS4ML-002DoneHLS4ML - Add trainable ModelGraph flow and validation passes
Done in hls4ml-trainable: registered vivado:trainable after template resolution and before writer emission, added vivado:validate_trainable_config for the first supported Dense trainable subset, and added vivado:resolve_trainable_backward_order to attach trainable_forward_path, trainable_forward_order, trainable_backward_order, and trainable_output_layer while rejecting branching graphs. Loss endpoint metadata remains tracked under HLS4ML-006.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-22
HLS4ML-003DoneHLS4ML - Add trainable layer attributes and precision resolution
Done in hls4ml-trainable: Layer initialization now writes a trainable flag and converts configured trainable precision fields into NamedType attributes such as grad_in_t, grad_out_t, raw_update_t, update_t, gradient_accum_t, controller_metric_t, and alpha_t.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-22
HLS4ML-004DoneHLS4ML - Port static trainable headers into Vivado templates
Done in hls4ml-trainable: templates/vivado/trainable is the new umbrella for common trace helpers, Dense backprop, MSE/half-MSE losses, SGD raw updates, and CTRL-NONE/global-throttle helpers. Generated firmware now copies this tree into firmware/trainable.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-22
HLS4ML-005DoneHLS4ML - Implement Dense backpass templates first
Done for the minimal path: nnet_dense_backprop.h computes dL/dx, accumulates dL/dW and dL/db across power-of-two batches, and emits averaged Dense gradients. Writer emission now allocates the required Dense backprop buffers.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-22
HLS4ML-006DoneHLS4ML - Generate loss endpoints and output-gradient seeds
Done for the first path: vivado:resolve_trainable_loss_endpoints attaches normalized one-output half_mse metadata, the trainable MSE header emits loss and dL/dy with shift-based scaling, and the Vivado writer emits loss config, dL/dy buffers, and loss kernel calls.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-22
HLS4ML-007DoneHLS4ML - Implement two-phase raw update storage for Dense
Done for Dense/CTRL-NONE: backprop emits averaged gradients, SGD emits raw update vectors, and apply_dense_update applies alpha-scaled updates in a separate phase. Global metric reductions remain under HLS4ML-009.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-22
HLS4ML-011DoneHLS4ML - Add minimal Vivado writer hooks for trainable emission
Done in hls4ml-trainable: the Vivado writer copies trainable headers, emits trainable includes/config structs, allocates trainable buffers, emits the minimal loss/backprop/SGD/CTRL-NONE/apply-update call chain, and keeps bridge inference calls safe with train_enable=false stubs.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-22
ENB-006CompletedImplement fake-fixed-point precision and diagnostic summaries
Added layer-indexed precision configuration, rail statistics, quantized update paths, and compact precision summaries.
- Due
- 2026-05-20
- Start
- 2026-05-15
- End
- 2026-05-20
ENB-005CompletedRefactor enabol neural training into controller-based modules
Split the old training loop into model, controller, optimizer, telemetry, applier, and trainer modules with Controller.from_str and MetricsConfig.
- Due
- 2026-05-20
- Start
- 2026-05-09
- End
- 2026-05-20
HLS4ML-038DoneHLS4ML - Emit controller metric trace from trainable CSIM
Generated trainable top-level functions now expose controller metric outputs and the trainable CSIM testbench writes tb_data/training/controller.dat. The current schema records raw_update_norm_sq, controlled_update_norm_sq, actual_update_norm_sq, dgrad_norm_sq, dtheta_for_control_sq, stability_lhs_raw, stability_lhs_ctrl, stability_rhs, alpha_feasible, alpha_state, alpha_code, alpha_min, and controller_feasible. TestbenchData loads controller.dat as top-level scalar traces.
- Due
- UNDEFINED
- Start
- 2026-05-23
- End
- 2026-05-23
ENB-024DoneAdd TestbenchData loader and plots for hls4ml trainable traces
Added TestbenchData.from_dir/from_trainable_dir for recursive tb_data/training/*.dat traces, per-trace metadata parsing, a table-style representation, sparse-cadence trace merging, namespaced nested trace columns, and TestbenchData.plot_training() for metadata, smoothed traces, global-step x-axis, and epoch top axis. FitHistory remains scoped to software training histories.
- Due
- UNDEFINED
- Start
- 2026-05-23
- End
- 2026-05-23
HLS4ML-020DoneHLS4ML - Validate minimal CTRL-NONE trainable compile path
Done: the one-Dense half_mse -> Dense backpass -> SGD -> CTRL-NONE -> apply-update path runs in CSIM on kona and produces training loss curves. Trainable projects emit config-sized local buffers, static learning rate in trainable_configN, an epoch/shuffle CSIM testbench, compact console progress, and tb_data/training traces.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-23
ENB-011DoneAdd multi-layer software benchmarks
Move beyond one-layer LIN1 into coupled dense models and nonlinear hidden representations before committing to hardware paths.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-22
ENB-009CompletedRun EXP-001 stability phase-diagram sweep
Generated the first software phase diagrams across learning-rate and drift parameters for controlled and uncontrolled training.
- Due
- 2026-05-21
- Start
- 2026-05-20
- End
- 2026-05-21
ENB-007CompletedRun EXP-000A floating-point global throttle sanity test
Verified the first one-layer software sanity case for global throttling under affine drift.
- Due
- 2026-05-19
- Start
- 2026-05-15
- End
- 2026-05-19
ENB-008CompletedRun EXP-000B fake-quantized global throttle sanity test
Validated staged fake-fixed-point paths and showed that throttling can recover in tight quantization regimes where no-controller training oscillates.
- Due
- 2026-05-20
- Start
- 2026-05-15
- End
- 2026-05-20
ENB-025DoneAdd per-layer parameter trace loading and statistics
Raw weights and biases now load into per-layer TestbenchLayerData objects instead of tb.frame. TestbenchData.from_dir(..., load_weights=True/False/[layers]) controls loading, and tb.stats_frame/tb.scalar_frame expose per-layer parameter summaries for plotting.
- Due
- UNDEFINED
- Start
- 2026-05-23
- End
- 2026-05-23
ENB-027DoneStandardize TestbenchData training panels for controller diagnostics
TestbenchData.plot_training now promotes expected training panels before generic scalar traces: loss, alpha, and a twin-axis controller-norm panel for ||ΔG|| and ||Δθ_raw||. The norms are derived from controller.dat squared accumulators, and callers can override y-axis scales with scales={metric: scale}.
- Due
- UNDEFINED
- Start
- 2026-05-23
- End
- 2026-05-23
ENB-012DoneCompare global throttle against naive safety mechanisms
Use clipping, loose κ projection, and legacy row/column projection as comparisons after the fixed-point path is understood.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-22
ENB-022DoneAdd scoped ENABOL toolchain profiles for hls4ml compilation
Added enabol.toolchain with profile loading, host/platform/backend validation, scoped environment application/restoration, required command checks, and tracked example profiles for the kona-ubuntu Vivado/Vitis installs. The default profile is kona-vitis-2024_1 because 2023.2 CSIM is not reliable on Ubuntu 24.04 without a container.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-22
HLS4ML-012DoneHLS4ML - Build the ENABOL to hls4ml trainable config bridge
Done for the config bridge: enabol.compile.compile builds hls4ml config, emits Model.Training, delegates PrecisionDict to hls4ml precision mapping to enabol.precision, applies toolchain profiles for build=True, exports dataset testbench files, and converts/writes/builds through hls4ml. Reference trace comparison continues under HLS4ML-013 once trainable kernels exist.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- 2026-05-22
ENB-026DoneRefactor curvature sensor from controller kernel for multi-layer global reduction
Split global_throttle_order0 into raw_update_sensor_order0 (per-layer raw update and ΔG squared-norm contribution) and global_throttle_order0_law (global law from accumulated norms). Writer emission: backpass, SGD raw proposal, raw-update/dgrad reduction, controller law, then scaled update per layer. GT-1 built on the same architecture.
- Due
- UNDEFINED
- Start
- 2026-05-23
- End
- 2026-05-23
Static Views
Unassigned Backlog
HLS4ML-025UnassignedHLS4ML - Reintroduce kappa safety budgets as optional bounds
Treat kappa as an optional safety bound or gain-rail budget, not as the primary learning throttle. Decide global versus per-layer kappa policy before adding hardware code.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-010UnassignedHLS4ML - Evaluate delayed-alpha pipelined update schedule
Analyze the one-step-delayed alpha schedule where alpha_t is computed from the current backpass but alpha_{t-1} is used to apply stored updates during the next forward traversal.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-023UnassignedHLS4ML - Implement CTRL-GT-ORDER-2 controller
Add second-order controller state after GT1. Validate alpha/velocity state update and fixed-point rails against software traces.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-024UnassignedHLS4ML - Implement CTRL-GT-ORDER-2-QA controller
Add the quantization-aware lower alpha bound once precision inference is available. This should use explicit update quantum metadata rather than ad hoc bit inspection.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-027UnassignedHLS4ML - Add binary crossentropy loss endpoints
Add binary_crossentropy and binary_crossentropy_from_logits metadata/templates, including final-sigmoid skip semantics and fixed-point/PWL helper review.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-028UnassignedHLS4ML - Add categorical crossentropy loss endpoints
Add categorical_crossentropy and categorical_crossentropy_from_logits, including final-softmax skip semantics and stable gradient seed policy.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-030UnassignedHLS4ML - Add activation backpasses
Port/rewrite activation backpasses for ReLU, sigmoid, tanh, softmax where appropriate. Start with the activation needed by the two-Dense validation model.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-036UnassignedHLS4ML - Add optimizer-state allocation framework
Generalize optimizer state allocation so SGD, Adam, and future optimizers do not hide static arrays inside layer kernels.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-026UnassignedHLS4ML - Add MAE loss endpoint support
Port or rewrite MAE loss and gradient seed after half_mse is validated. Define zero-difference subgradient policy explicitly.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-029UnassignedHLS4ML - Add secondary regression/divergence losses
Future support for MAPE, MSLE, cosine similarity, and KL divergence. These should not block the first Dense trainable compile path.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-031UnassignedHLS4ML - Add BatchNorm backpass
Add BatchNorm backward support after Dense and activation paths are stable. Reconcile old BatchNorm backprop code with current hls4ml BatchNormalization layer metadata.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-032UnassignedHLS4ML - Add Conv1D and Conv2D backpasses
Add convolution backpasses only after the Dense path proves the graph-flow/template interface. Start from Conv1D or the smallest Conv2D fixture, not from the full deprecated generic code.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-033UnassignedHLS4ML - Add pooling backpasses
Future support for max/average/global pooling gradients. This likely needs cached argmax or mask policy for max pooling.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-034UnassignedHLS4ML - Add layout-only backpasses
Port/rewrite simple layout gradients such as Flatten, Reshape, ZeroPadding, Cropping, and UpSampling after activation traversal is working.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-035UnassignedHLS4ML - Add Adam optimizer support
Future Adam support with explicit state buffers and bias-correction policy. The deprecated code mixes double/std::pow paths and large static arrays, so this must be rewritten carefully before synthesis.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-013UnassignedHLS4ML - Validate one-layer Dense trainable CSIM
Run the minimal one-dense affine regression case through generated trainable firmware in CSIM and compare forward values, loss, gradients, alpha, and updated weights against ENABOL software traces.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
ENB-017UnassignedRun CSIM validation with global throttle (CTRL-GT-ORDER-0, division-free fix)
GT-0 CSIM failed: two ap_fixed divisions (sqrt + divide) crashed CSIM at step 1. Fixed by replacing α = χ/(ηC+ε) with division-free inequality search α²·η²·||ΔG||² ≤ χ²·(||Δθ_raw||² + ε²). The raw update is the optimizer proposal before alpha/quantized assignment; for SGD it already includes η. Candidate table of binary-fraction α values is searched in descending order with a nonzero alpha floor. Validate: compile 1-Dense model with Kind: CTRL-GT-ORDER-0, run CSIM, confirm α selection, no CSIM errors, loss drops.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-014UnassignedHLS4ML - Extend trainable CSIM to two Dense layers with activation
After the one-layer Dense path is correct, validate reverse traversal through Dense -> activation -> Dense before moving to convolution and BatchNorm paths.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
ENB-023UnassignedInfer hls4ml trainable precisions from ENABOL model, dataset, optimizer, and controller
Replace placeholder/default trainable precision fields with an inference pass that uses model ranges, dataset ranges, learning rate, batch size, optimizer state, controller metrics, loss scale, and fixed-point safety margins.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
ENB-019UnassignedRun FPGA hardware drift-recovery experiment
Collect the real hardware baseline, post-drift initial point, and post-training recovery point for controller-enabled learning.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
ENB-018UnassignedRun HLS synthesis and resource validation
Check resource impact, timing, and generated design viability after CSIM validates behavior.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
ENB-021UnassignedPrepare paper narrative, plots, and handoff package
Turn the final validated story into paper-ready figures, narrative, and handoff documentation.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
In Progress
ENB-013In progressCreate reusable precision-map records
Promote repeated fixed-point formats into stable PREC-* records once multiple experiments reuse the same precision maps.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- UNDEFINED
HLS4ML-008In progressHLS4ML - Implement synthesizable controller kernels
CTRL-NONE, CTRL-GT-ORDER-0, and CTRL-GT-ORDER-1 are implemented and wired. GT-0/GT-1 now use raw optimizer update geometry plus division-free candidate search over α²·η²·||ΔG||² ≤ χ²·(||Δθ_raw||²+ε²), with a nonzero alpha candidate floor. CTRL-GT-ORDER-2/2-QA remain.
- Due
- UNDEFINED
- Start
- 2026-05-22
- End
- UNDEFINED
HLS4ML-009In progressHLS4ML - Add global metric reductions and alpha synchronization
Raw-update controller sensing is separated from controller law (raw_update_sensor_order0 → global_throttle_*_law). The writer emission now runs backpass → SGD proposal → raw-update/dgrad reduction → law → alpha-scaled apply. Per-layer raw update norms and ΔG norms are summed into global accumulators before the law call. CTRL-NONE skips curvature sensing.
- Due
- UNDEFINED
- Start
- UNDEFINED
- End
- UNDEFINED
HLS4ML-022In progressHLS4ML - Implement CTRL-GT-ORDER-1 controller
Implemented global_throttle_order1_law with persistent alpha_state. Uses the same raw-update controller geometry and writer emission as GT-0. KAlpha config field added to bridge and config struct. Pending CSIM validation after the raw-update geometry fix.
- Due
- UNDEFINED
- Start
- 2026-05-23
- End
- UNDEFINED
ENB-010In progressExpand software stress tests beyond the first sweep
Run distribution shift, transient spike, rail saturation, and update dead-zone tests across enough benchmarks to stress the controller.
- Due
- UNDEFINED
- Start
- 2026-05-20
- End
- UNDEFINED
HLS4ML-037In progressHLS4ML - Add optional trainable CSIM trace logging
Current slice implemented: generated trainable CSIM writes metadata-bearing comma-delimited .dat traces under tb_data/training/ for loss, alpha, and per-layer epoch-level parameter traces such as dense0/weights.dat and dense0/biases.dat. Repeated epoch/sample/global_step/sample_index columns allow sparse traces with different cadences. Broader trace selection for gradients and controller internals remains open.
- Due
- UNDEFINED
- Start
- 2026-05-23
- End
- UNDEFINED
ENB-014In progressValidate Conv2D backprop path in hls4ml trainable backend
Re-run MSWC-style tests with trainable convolutional layers, checking gradient flow through frozen dense layers and correctness of conv2d backpass.
- Due
- 2026-05-25
- Start
- 2026-05-18
- End
- UNDEFINED
ENB-015In progressImplement and validate BatchNorm backprop
Correct BN backprop even when BN parameters are frozen, because gradients must propagate through BN into earlier trainable layers.
- Due
- 2026-05-25
- Start
- 2026-05-18
- End
- UNDEFINED
ENB-016In progressFork hls4ml:main, create :trainable or :enabol branch and set up repo
Fork the hls4ml repo and set up the branch structure and CI paths for integrating the trainable controller code path into hls4ml templates and vivado writer.
- Due
- 2026-05-28
- Start
- 2026-05-21
- End
- 2026-05-21
ENB-016bIn progressAnalyze the hls4ml repo and trace an updated strategy for integrating our controller.
Port the selected global throttle path from software experiments into the hls4ml trainable code path and code generation templates.
- Due
- 2026-05-28
- Start
- 2026-05-21
- End
- UNDEFINED
ENB-020In progressMaintain project meeting notes and task status
Keep the project memory current as the work shifts from software ablations to hls4ml and hardware.
- Due
- UNDEFINED
- Start
- 2026-05-20
- End
- UNDEFINED
Blocked
No tasks match this filter.
Completed Tasks
| ID | Task | Status | Priority | Stage | Target | Action | Owner | Due | Start | End | Links |
|---|---|---|---|---|---|---|---|---|---|---|---|
ENB-001 | Build the ENABOL documentation and project registry system Created the Docusaurus structure, metadata components, people registry, task board, meeting calendar, and implementation/experiment hubs. | Completed | High | Research / theory | Documentation site | Documentation | @Manu B. Valentin | 2026-05-19 | 2026-05-1 | 2026-05-16 | Docs HomeCode Map |
ENB-003 | Study global throttle Separated static κ-style representational rails from dynamic trajectory control and documented why global throttling preserves update direction. | Completed | High | Research / theory | Theory | Analysis | @Manu B. Valentin | 2026-05-20 | 2026-05-09 | 2026-05-18 | Global ThrottleHow This Fits |
ENB-002 | Study κ-budgeting failure mode Separated static κ-style representational rails from dynamic trajectory control and documented why global throttling preserves update direction. | Completed | High | Research / theory | Theory | Analysis | @Manu B. Valentin | 2026-05-09 | 2026-05-01 | 2026-05-09 | Global ThrottleHow This Fits |
ENB-028 | Document trainable precision schema relationships Added the first precision relationship document. Unknown trainable precision dependencies are explicitly marked with ?, and the known rule for controller_metric records that dtheta_sq needs fractional bits for squared parameter movement: F_controller_metric >= 2*F_theta + guard bits. | Done | High | System design | precision | Documentation | @Manu B. Valentin | UNDEFINED | 2026-05-23 | 2026-05-23 | Trainable Precision Schemaprecision.pyPhase 1 Trainable Config |
HLS4ML-001 | HLS4ML - Define the trainable HLSConfig schema Done in hls4ml-trainable: HLSConfig now has normalized Model.Training defaults and accessors for trainable mode, loss, optimizer, controller, trainable precision, recognized precision fields, and per-layer Training overrides. | Done | High | System design | hls4ml | Design | @Manu B. Valentin | UNDEFINED | 2026-05-21 | 2026-05-22 | HLSConfighls4ml IntroPhase 1 Config |
ENB-004 | Define the controller registry and stability stress-test catalog Established CTRL-* pages and ST-* stress-test pages as stable IDs for experiments and task tracking. | Completed | High | System design | Controller stack | Design | @Manu B. Valentin | 2026-05-20 | 2026-05-09 | 2026-05-18 | ControllersAblations |
HLS4ML-039 | HLS4ML - Use raw-update geometry for global throttle control Replaced actual/throttled parameter-delta controller geometry with raw optimizer update geometry. SGD now runs before the law, raw_update_sensor_order0 sums ||Δθ_raw||² and ||ΔG||² globally, GT-0/GT-1 keep the division-free squared inequality, alpha uses a nonzero candidate floor, and controller.dat logs raw/controlled/actual update norms plus feasibility telemetry. | Done | High | Implementation | Controller stack | Development | @Manu B. Valentin | UNDEFINED | 2026-05-23 | 2026-05-23 | GT-0GT-1Vivado Writertestbench.py |
HLS4ML-015 | HLS4ML - Port minimal trainable static headers Done: minimal static support files now live under templates/vivado/trainable/{common,losses,backprop,optimizers,controllers}. This includes trainable_trace.h, mse.h, nnet_dense_backprop.h, sgd.h, and global_throttle.h. | Done | High | Implementation | C++ backend | Development | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-22 | Template PassesLoss Endpoints |
HLS4ML-016 | HLS4ML - Implement half-MSE loss template and dLdy seed Done: mse.h implements mse and half_mse using one shift-based core, with half_mse seeding dL/dy = y_hat - y and mse seeding dL/dy = 2 * (y_hat - y). | Done | High | Implementation | C++ backend | Development | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-22 | Loss Endpoints |
HLS4ML-017 | HLS4ML - Emit trainable top-level IO for loss and training controls Done for the minimal path: generated top-level firmware receives ground truth, scalar loss output, alpha output, train_enable, reset_accumulators, and batch_end. The testbench trains from tb_output_predictions.dat, while the Python bridge supplies zero trainable tensors with train_enable=false for inference-safe calls. | Done | High | Implementation | hls4ml | Development | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-22 | Vivado WriterLoss Endpoints |
HLS4ML-018 | HLS4ML - Implement minimal Dense backpass without controller Done: the first Dense backward kernel computes dL/dx, dL/dW, and dL/db accumulation and the writer emits the corresponding call for the resolved backward traversal. Global throttle, kappa, Adam, and delayed-alpha scheduling remain future work. | Done | High | Implementation | C++ backend | Development | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-22 | Template Passes |
HLS4ML-019 | HLS4ML - Implement raw SGD update proposal for Dense Done: sgd.h computes raw weight/bias update proposals from averaged gradients and learning rate, and writer emission calls SGD before CTRL-NONE alpha application. | Done | High | Implementation | C++ backend | Development | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-22 | Optimizer + Controller |
HLS4ML-021 | HLS4ML - Implement CTRL-GT-ORDER-0 controller Implemented CTRL-GT-ORDER-0 as a global law fed by raw_update_sensor_order0. The writer sums per-layer ||Δθ_raw||² and ||ΔG||² contributions into global accumulators, then GT-0 selects the largest candidate alpha satisfying α²·η²·||ΔG||² ≤ χ²·(||Δθ_raw||²+ε²). This avoids sqrt/division and avoids the self-locking actual-update denominator. | Done | High | Implementation | Controller stack | Development | @Manu B. Valentin | UNDEFINED | 2026-05-23 | 2026-05-23 | ControllersGlobal Throttle |
HLS4ML-002 | HLS4ML - Add trainable ModelGraph flow and validation passes Done in hls4ml-trainable: registered vivado:trainable after template resolution and before writer emission, added vivado:validate_trainable_config for the first supported Dense trainable subset, and added vivado:resolve_trainable_backward_order to attach trainable_forward_path, trainable_forward_order, trainable_backward_order, and trainable_output_layer while rejecting branching graphs. Loss endpoint metadata remains tracked under HLS4ML-006. | Done | High | Implementation | hls4ml | Development | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-22 | ModelGraph + Flows |
HLS4ML-003 | HLS4ML - Add trainable layer attributes and precision resolution Done in hls4ml-trainable: Layer initialization now writes a trainable flag and converts configured trainable precision fields into NamedType attributes such as grad_in_t, grad_out_t, raw_update_t, update_t, gradient_accum_t, controller_metric_t, and alpha_t. | Done | High | Implementation | hls4ml | Development | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-22 | Layer AttributesPrecisionPhase 1 Config |
HLS4ML-004 | HLS4ML - Port static trainable headers into Vivado templates Done in hls4ml-trainable: templates/vivado/trainable is the new umbrella for common trace helpers, Dense backprop, MSE/half-MSE losses, SGD raw updates, and CTRL-NONE/global-throttle helpers. Generated firmware now copies this tree into firmware/trainable. | Done | High | Implementation | C++ backend | Development | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-22 | Template Passes |
HLS4ML-005 | HLS4ML - Implement Dense backpass templates first Done for the minimal path: nnet_dense_backprop.h computes dL/dx, accumulates dL/dW and dL/db across power-of-two batches, and emits averaged Dense gradients. Writer emission now allocates the required Dense backprop buffers. | Done | High | Implementation | C++ backend | Development | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-22 | Template Passes |
HLS4ML-006 | HLS4ML - Generate loss endpoints and output-gradient seeds Done for the first path: vivado:resolve_trainable_loss_endpoints attaches normalized one-output half_mse metadata, the trainable MSE header emits loss and dL/dy with shift-based scaling, and the Vivado writer emits loss config, dL/dy buffers, and loss kernel calls. | Done | High | Implementation | hls4ml | Development | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-22 | Loss Endpoints |
HLS4ML-007 | HLS4ML - Implement two-phase raw update storage for Dense Done for Dense/CTRL-NONE: backprop emits averaged gradients, SGD emits raw update vectors, and apply_dense_update applies alpha-scaled updates in a separate phase. Global metric reductions remain under HLS4ML-009. | Done | High | Implementation | C++ backend | Development | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-22 | Optimizer + Controller |
HLS4ML-011 | HLS4ML - Add minimal Vivado writer hooks for trainable emission Done in hls4ml-trainable: the Vivado writer copies trainable headers, emits trainable includes/config structs, allocates trainable buffers, emits the minimal loss/backprop/SGD/CTRL-NONE/apply-update call chain, and keeps bridge inference calls safe with train_enable=false stubs. | Done | High | Implementation | hls4ml | Development | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-22 | Vivado Writer |
ENB-006 | Implement fake-fixed-point precision and diagnostic summaries Added layer-indexed precision configuration, rail statistics, quantized update paths, and compact precision summaries. | Completed | High | Implementation | enabol module | Development | @Manu B. Valentin | 2026-05-20 | 2026-05-15 | 2026-05-20 | PrecisionQuantization |
ENB-005 | Refactor enabol neural training into controller-based modules Split the old training loop into model, controller, optimizer, telemetry, applier, and trainer modules with Controller.from_str and MetricsConfig. | Completed | High | Implementation | enabol module | Development | @Manu B. Valentin | 2026-05-20 | 2026-05-09 | 2026-05-20 | NN ModuleTraining Loop |
HLS4ML-038 | HLS4ML - Emit controller metric trace from trainable CSIM Generated trainable top-level functions now expose controller metric outputs and the trainable CSIM testbench writes tb_data/training/controller.dat. The current schema records raw_update_norm_sq, controlled_update_norm_sq, actual_update_norm_sq, dgrad_norm_sq, dtheta_for_control_sq, stability_lhs_raw, stability_lhs_ctrl, stability_rhs, alpha_feasible, alpha_state, alpha_code, alpha_min, and controller_feasible. TestbenchData loads controller.dat as top-level scalar traces. | Done | High | Validation / benchmarking | CSIM | Development | @Manu B. Valentin | UNDEFINED | 2026-05-23 | 2026-05-23 | Phase 5 Bridge + Validationtestbench.py |
ENB-024 | Add TestbenchData loader and plots for hls4ml trainable traces Added TestbenchData.from_dir/from_trainable_dir for recursive tb_data/training/*.dat traces, per-trace metadata parsing, a table-style representation, sparse-cadence trace merging, namespaced nested trace columns, and TestbenchData.plot_training() for metadata, smoothed traces, global-step x-axis, and epoch top axis. FitHistory remains scoped to software training histories. | Done | High | Validation / benchmarking | CSIM | Development | @Manu B. Valentin | UNDEFINED | 2026-05-23 | 2026-05-23 | history.pytestbench.pyPhase 5 Bridge + Validation |
HLS4ML-020 | HLS4ML - Validate minimal CTRL-NONE trainable compile path Done: the one-Dense half_mse -> Dense backpass -> SGD -> CTRL-NONE -> apply-update path runs in CSIM on kona and produces training loss curves. Trainable projects emit config-sized local buffers, static learning rate in trainable_configN, an epoch/shuffle CSIM testbench, compact console progress, and tb_data/training traces. | Done | High | Validation / benchmarking | CSIM | Testing | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-23 | Phase 5 Bridge + Validation |
ENB-011 | Add multi-layer software benchmarks Move beyond one-layer LIN1 into coupled dense models and nonlinear hidden representations before committing to hardware paths. | Done | High | Validation / benchmarking | Software experiments | Experiment | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-22 | Experiments |
ENB-009 | Run EXP-001 stability phase-diagram sweep Generated the first software phase diagrams across learning-rate and drift parameters for controlled and uncontrolled training. | Completed | High | Validation / benchmarking | Software experiments | Experiment | @Manu B. Valentin | 2026-05-21 | 2026-05-20 | 2026-05-21 | EXP-001 |
ENB-007 | Run EXP-000A floating-point global throttle sanity test Verified the first one-layer software sanity case for global throttling under affine drift. | Completed | High | Validation / benchmarking | Software experiments | Experiment | @Manu B. Valentin | 2026-05-19 | 2026-05-15 | 2026-05-19 | EXP-000A |
ENB-008 | Run EXP-000B fake-quantized global throttle sanity test Validated staged fake-fixed-point paths and showed that throttling can recover in tight quantization regimes where no-controller training oscillates. | Completed | High | Validation / benchmarking | Software experiments | Experiment | @Manu B. Valentin | 2026-05-20 | 2026-05-15 | 2026-05-20 | EXP-000B |
ENB-025 | Add per-layer parameter trace loading and statistics Raw weights and biases now load into per-layer TestbenchLayerData objects instead of tb.frame. TestbenchData.from_dir(..., load_weights=True/False/[layers]) controls loading, and tb.stats_frame/tb.scalar_frame expose per-layer parameter summaries for plotting. | Done | Medium | Validation / benchmarking | CSIM | Development | @Manu B. Valentin | UNDEFINED | 2026-05-23 | 2026-05-23 | testbench.pyPhase 5 Bridge + Validation |
ENB-027 | Standardize TestbenchData training panels for controller diagnostics TestbenchData.plot_training now promotes expected training panels before generic scalar traces: loss, alpha, and a twin-axis controller-norm panel for ||ΔG|| and ||Δθ_raw||. The norms are derived from controller.dat squared accumulators, and callers can override y-axis scales with scales={metric: scale}. | Done | Medium | Validation / benchmarking | plots | Development | @Manu B. Valentin | UNDEFINED | 2026-05-23 | 2026-05-23 | testbench.pyPhase 5 Bridge + Validation |
ENB-012 | Compare global throttle against naive safety mechanisms Use clipping, loose κ projection, and legacy row/column projection as comparisons after the fixed-point path is understood. | Done | Medium | Validation / benchmarking | Software experiments | Analysis | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-22 | ST-006ST-007 |
ENB-022 | Add scoped ENABOL toolchain profiles for hls4ml compilation Added enabol.toolchain with profile loading, host/platform/backend validation, scoped environment application/restoration, required command checks, and tracked example profiles for the kona-ubuntu Vivado/Vitis installs. The default profile is kona-vitis-2024_1 because 2023.2 CSIM is not reliable on Ubuntu 24.04 without a container. | Done | High | Integration | compile | Development | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-22 | Toolchain ProfilesPhase 5 Bridge + Validation |
HLS4ML-012 | HLS4ML - Build the ENABOL to hls4ml trainable config bridge Done for the config bridge: enabol.compile.compile builds hls4ml config, emits Model.Training, delegates PrecisionDict to hls4ml precision mapping to enabol.precision, applies toolchain profiles for build=True, exports dataset testbench files, and converts/writes/builds through hls4ml. Reference trace comparison continues under HLS4ML-013 once trainable kernels exist. | Done | High | Integration | enabol module | Integration | @Manu B. Valentin | UNDEFINED | 2026-05-22 | 2026-05-22 | ENABOL BridgePhase 5 Bridge + Validation |
ENB-026 | Refactor curvature sensor from controller kernel for multi-layer global reduction Split global_throttle_order0 into raw_update_sensor_order0 (per-layer raw update and ΔG squared-norm contribution) and global_throttle_order0_law (global law from accumulated norms). Writer emission: backpass, SGD raw proposal, raw-update/dgrad reduction, controller law, then scaled update per layer. GT-1 built on the same architecture. | Done | High | refactoring | C++ backend | Development | @Manu B. Valentin | UNDEFINED | 2026-05-23 | 2026-05-23 | ControllersGlobal Throttle |
Useful Filters
<TaskBoard status="todo" />
<TaskBoard owner="__unassigned__" />
<TaskBoard owner="mbvalentin" />
<TaskBoard stage="validation" />
<TaskBoard target="enabol" />
<TaskBoard action="development" />
<TaskBoard tag="global-throttle" />
<TaskBoard statuses={["todo", "inprogress"]} view="labels" />