Dev/molt feature dashboard#23
Open
Ky-Ng wants to merge 4 commits into
Open
Conversation
Adds the Molt model class, MoltModule Lightning wrapper, and the necessary plumbing to support a single-layer MoLT alongside the existing CrossLayerTranscoder. Includes: - crosslayer_transcoder/model/molt.py: new Molt nn.Module - model/__init__.py: export Molt - model/jumprelu.py: allow n_layers=1 to produce a 2-D theta parameter - model/clt_lightning.py: import Molt, widen model type to Union[CrossLayerTranscoder, Molt], wrap the encoder/decoder assertions and last_active buffer in an isinstance check, and append the MoltModule subclass with its own training_step - data/datamodule.py: guard self.data_loader teardown with is not None Known limitations (follow-ups): - MoltModule.training_step is hardcoded to layer 8 - compute_dead_features config flags are inert for MoLT - Molt does not yet inherit from SerializableModule / save_pretrained Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Eight YAML configs for the Lightning CLI: - config/molt.yaml, config/molt-long.yaml: baseline - config/molt-5090.yaml: tuned for a 5090 with 31 GB /dev/shm - config/molt-5090_20M_tokens_*.yaml: sparsity sweep at 20M tokens - config/molt-5090_50M_tokens_0_00015.yaml: 50M-token run class_path entries point at the master package (crosslayer_transcoder.*). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
tests/test_molt_smoke.py covers three cases: - test_molt_cpu_forward: builds a tiny Molt, runs one forward pass, checks shapes and finiteness — no GPU, no Lightning, no dataset - test_molt_gpu_fp32_train_step: forward + backward + Adam step on synthetic activations on cuda; asserts loss and params remain finite - test_molt_gpu_amp_train_step: same, inside torch.amp.autocast(float16) with a GradScaler — mirrors Lightning's precision="16-mixed" Both GPU tests are guarded with skipif(not cuda.is_available()), so they silently skip on CPU CI runners while still exercising mixed-precision locally. Verified locally on RTX 5090: 3/3 tests pass; full suite is 210 passed. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.