left overs

config/trainer.yaml

@@ -70,7 +70,7 @@ training:
     batch_num_samples: 16
     grad_acc_steps: 1
     max_grad_norm: 10.0
-    start_after_epochs: 1500
+    start_after_epochs: 300
     steps_per_epoch: 40
     imagine_horizon: ${common.sequence_length}
     burn_in: 20

justfile

@@ -6,8 +6,8 @@ breakout:
 crafter:
     python src/main.py env.train.id=CrafterReward-v1
 
-minihack:
-    python src/main.py env.train.id=MiniHack-River-v0
+# minihack:
+#     python src/main.py env.train.id=MiniHack-River-v0
 
 # watch the latest runs
 watch_latest:
@@ -17,5 +17,13 @@ watch_latest:
     cd *([-1]) && \
     scripts/play.sh -e -r -h
 
+
+resume_latest:
+    . ./.venv/bin/activate
+    cd ./outputs && \
+    cd *([-1]) && \
+    cd *([-1]) && \
+    scripts/resume.sh
+
 default: 
     just --list

research_journal.md

@@ -375,7 +375,7 @@ So IRIS has
     ```
 - Dynamics $D(z_0, a_0) = z_1$
     ```py
-    outputs_wm = self.world_model(token, past_keys_values=self.keys_values_wm)
+    outputs_wm = self.world_model(tokenRedmond AI, past_keys_values=self.keys_values_wm)
     ```
 - Decoder $D(z_0, a_0) = x_1$
     ```py
@@ -400,3 +400,8 @@ To summarize
 
 
 no it didn't work with tokenizer sep hmm
+
+Oh it did with whole transfrmer and tokenizer at same time https://wandb.ai/wassname/iris/runs/w7lvs4gi?workspace=user-wassname
+wandb:           world_model/eval/loss_obs ▇█▃▄▄▄▄▃▃▂▂▂▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁
+wandb:       world_model/eval/loss_rewards ▁█▃█▇▅▇▁▅▆▂▄▃▇▆▃▄▇▃▄▂▄▃▂▃▄▃
+wandb:         world_model/eval/total_loss ▂█▃██▅▇▁▅▆▁▃▂█▇▃▃▆▃▄▂▃▃▂▂▄▂