hierarchical-decision-transformer-hopper

README.md

@@ -38,7 +38,7 @@ The following hyperparameters were used during training:
 - optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
 - lr_scheduler_type: linear
 - lr_scheduler_warmup_ratio: 0.1
-- num_epochs: 200
+- num_epochs: 120
 
 ### Training results
 

config.json

@@ -1,12 +1,13 @@
 {
-  "act_dim": 6,
+  "act_dim": 3,
   "action_tanh": true,
   "activation_function": "relu",
   "architectures": [
-    "TrainableDT"
+    "HierarchicalTrainableDT"
   ],
   "attn_pdrop": 0.1,
   "bos_token_id": 50256,
+  "cluster_hidden_dim": 128,
   "embd_pdrop": 0.1,
   "eos_token_id": 50256,
   "hidden_size": 128,
@@ -14,6 +15,7 @@
   "layer_norm_epsilon": 1e-05,
   "max_ep_len": 4096,
   "model_type": "decision_transformer",
+  "n_clusters": 8,
   "n_head": 1,
   "n_inner": null,
   "n_layer": 3,
@@ -22,9 +24,10 @@
   "resid_pdrop": 0.1,
   "scale_attn_by_inverse_layer_idx": false,
   "scale_attn_weights": true,
-  "state_dim": 17,
+  "state_dim": 11,
   "torch_dtype": "float32",
   "transformers_version": "4.46.2",
   "use_cache": true,
+  "use_subgoal_weighting": true,
   "vocab_size": 1
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4f9fefae1efcb89df5ef5bdc1b71b5ace0041243c894f08e3dbb7fa8995997a2
-size 5034448
+oid sha256:b1594b23cce583700d3ad77126a99316cd24a114cb910cfa0c02403861153946
+size 5036908

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:88c3d511ad5dfde27e4489c7390c4d7926683ad0a7458b50c4ebfc8f44f1f29c
+oid sha256:7b450a94cc940aa7ec181d60e07338eced46b079a4c4d822ad6444a2bd91a472
 size 5240