Weight persistence module #46

New Issue

2026-08-20T23:28:35+02:00

SirStone commented

2026-08-20 23:28:35 +02:00

Parent

What to build

Save and load all network tensors (actor, 2 critics, 2 target critics, alpha parameter, Adam optimizer states) as individual .npy tensors packed into a single .zip file via Nim's std/zipfiles.

Each tensor is a standard .npy inside the archive — individually inspectable by extracting. Atomic save: write to a temp file, then rename to the target path.

Checkpoint directory convention: weights/latest.zip for current state, periodic snapshots as weights/checkpoint_<N>.zip.

Acceptance criteria

save(path, networks) creates a valid .zip containing .npy entries for all tensors
load(path) → networks restores all tensors exactly (round-trip correctness)
File is a valid zip — can be extracted with standard tools and .npy files are readable by numpy
Atomic save: incomplete writes don't corrupt the checkpoint
test_weights.nim passes

Blocked by

#38 (SAC_LSTM_Bot: project scaffold)
#41 (LSTM network module)

## Parent #37 ## What to build Save and load all network tensors (actor, 2 critics, 2 target critics, alpha parameter, Adam optimizer states) as individual `.npy` tensors packed into a single `.zip` file via Nim's `std/zipfiles`. Each tensor is a standard `.npy` inside the archive — individually inspectable by extracting. Atomic save: write to a temp file, then rename to the target path. Checkpoint directory convention: `weights/latest.zip` for current state, periodic snapshots as `weights/checkpoint_<N>.zip`. ## Acceptance criteria - [ ] `save(path, networks)` creates a valid `.zip` containing `.npy` entries for all tensors - [ ] `load(path) → networks` restores all tensors exactly (round-trip correctness) - [ ] File is a valid zip — can be extracted with standard tools and `.npy` files are readable by numpy - [ ] Atomic save: incomplete writes don't corrupt the checkpoint - [ ] `test_weights.nim` passes ## Blocked by - #38 (SAC_LSTM_Bot: project scaffold) - #41 (LSTM network module)

SirStone added the ready-for-agent label 2026-08-20 23:28:35 +02:00

SirStone referenced this issue

2026-08-20 23:28:46 +02:00

SAC training module #47

SirStone referenced this issue

2026-08-20 23:29:01 +02:00

Main bot integration #48

SirStone commented

2026-08-20 23:59:36 +02:00

Weight persistence module complete — zipped .npy via std/zipfiles, atomic save (temp+rename), checkpoint includes Adam states with optional load. 12/12 tests pass. Full suite 69+ tests all green. Code reviewed clean.

SirStone closed this issue

2026-08-20 23:59:36 +02:00

SirStone referenced this issue from a commit

2026-08-24 09:05:45 +02:00

feat(SAC_LSTM_Bot): weight persistence module (#46)

SirStone referenced this issue from a commit

2026-08-24 09:05:45 +02:00

Merge branch 'worktree-agent-a8622248' (ticket #46 weight persistence)

Sign in to join this conversation.