From 66e0d8b9fd4d0f7a2231d689c055e26fdf1cf04a Mon Sep 17 00:00:00 2001 From: YurenHao0426 Date: Sat, 13 Jun 2026 12:35:36 -0500 Subject: rrm workspace: TRM/HRM/SRM code, Maze dataset, dynamical-analysis pipeline Curated export for clone-and-run Maze training (2x A6000) + diagnostics. trm/hrm pretrain.py carry trajectory-augmentation code (backward-compatible). Heavy artifacts (checkpoints/wandb/npz) gitignored; see PROVENANCE.md. Co-Authored-By: Claude Fable 5 --- trm/config/arch/trm_hier6.yaml | 26 ++++++++++++++++++++++++++ 1 file changed, 26 insertions(+) create mode 100644 trm/config/arch/trm_hier6.yaml (limited to 'trm/config/arch/trm_hier6.yaml') diff --git a/trm/config/arch/trm_hier6.yaml b/trm/config/arch/trm_hier6.yaml new file mode 100644 index 0000000..43622ec --- /dev/null +++ b/trm/config/arch/trm_hier6.yaml @@ -0,0 +1,26 @@ +name: recursive_reasoning.trm_hier6@TinyRecursiveReasoningModel_ACTV1 +loss: + name: losses@ACTLossHead + loss_type: stablemax_cross_entropy + +halt_exploration_prob: 0.1 +halt_max_steps: 16 + +H_cycles: 3 +L_cycles: 6 + +H_layers: 0 +L_layers: 2 + +hidden_size: 512 +num_heads: 8 # min(2, hidden_size // 64) +expansion: 4 + +puzzle_emb_ndim: ${.hidden_size} + +pos_encodings: rope +forward_dtype: bfloat16 + +mlp_t: False # use mlp on L instead of transformer +puzzle_emb_len: 16 # if non-zero, its specified to this value +no_ACT_continue: True # No continue ACT loss, only use the sigmoid of the halt which makes much more sense \ No newline at end of file -- cgit v1.2.3