Upload config.yaml with huggingface_hub

config.yaml

@@ -1,6 +1,6 @@
 logging:
     project: titok_video
-    run_name: BaseAll-CB16k-TL128-256x33-init-BS64-clipGrad1.0
+    run_name: BaseAll-CB16k-TL64-128x17-BS256
 
     logging_interval: 50
 
@@ -9,7 +9,7 @@ logging:
     keep_prior_checkpoints: -1 # -1 to keep all
 
     resume_from_checkpoint:
-    init_from_checkpoint: base-interp-256x33-TL128.ckpt
+    init_from_checkpoint:
 
 model:
     titok:
@@ -17,7 +17,7 @@ model:
         spatial_patch_size: 4
 
         fsq_levels: [8, 8, 8, 6, 5] # [7, 5, 5, 5, 5]
-        num_latent_tokens: 128
+        num_latent_tokens: 64
 
         encoder_size: base
         decoder_size: base
@@ -49,10 +49,10 @@ model:
         final_gamma: 0.1
 
 dataset:
-    train_dataset: "/workspace/out_enc_256_33/**/*.pt"
-    eval_dataset: "/workspace/out_enc_256_33_eval/*.pt"
-    resolution: 256
-    num_frames: 33
+    train_dataset: "/workspace/out_enc_128_17/**/*.pt"
+    eval_dataset: "/workspace/out_enc_128_17_eval/*.pt"
+    resolution: 128
+    num_frames: 17
     frames_per_second: 8
     workers: 8
 
@@ -62,7 +62,7 @@ optimizer:
         beta1: 0.9
         beta2: 0.99
         weight_decay: 1e-4
-        warmup_steps: 5000 # 10000
+        warmup_steps: 10000
         end_lr: 1e-5
 
     disc: # not used
@@ -76,9 +76,9 @@ optimizer:
 training:
     torch_compile: True
     seed: 42
-    max_grad_norm: 1.0 # not needed?
+    max_grad_norm: # 1.0
     
-    batch_size: 64
+    batch_size: 256
     # strategy: # ddp
 
     enable_tf32: True