Upload folder using huggingface_hub

output/two_stage_encoding/config.yaml

@@ -0,0 +1,137 @@
+out_dir: output/two_stage_encoding
+seed: 3315
+overwrite: false
+device: cuda
+batch_size: 16
+stage1:
+  epochs: 10
+  lr: 0.0003
+  weight_decay: 0.1
+  model:
+    embed_dim: 192
+    encoder_kernel_size: 45
+    decoder_kernel_size: 0
+    hidden_model: null
+    global_pool: avg
+    encoder_causal: false
+    encoder_positive: false
+    encoder_blockwise: false
+    pool_num_heads: 3
+    with_shared_decoder: true
+    with_subject_decoders: true
+  transformer:
+    num_heads: 3
+    depth: 6
+    mlp_ratio: 4.0
+  conv1dnext:
+    depth: 6
+    kernel_size: 11
+    causal: false
+stage2:
+  epochs: 10
+  lr: 0.0001
+  weight_decay: 0.01
+  cfm:
+    solver: euler
+    sigma_min: 0.0001
+  decoder:
+    channels:
+    - 256
+    - 256
+    dropout: 0.05
+    attention_head_dim: 64
+    n_blocks: 1
+    num_mid_blocks: 2
+    num_heads: 2
+    act_fn: snakebeta
+    down_block_type: transformer
+    mid_block_type: transformer
+    up_block_type: transformer
+subjects:
+- 1
+- 2
+- 3
+- 5
+features:
+  internvl3_8b:
+    model: InternVL3_8B
+    layers:
+      layers.20: language_model.model.layers.20.post_attention_layernorm
+  internvl3_14b:
+    model: InternVL3_14B
+    layers:
+      layers.20: language_model.model.layers.20.post_attention_layernorm
+      layers.30: language_model.model.layers.30.post_attention_layernorm
+  qwen-2-5-omni-3b:
+    model: qwen2-5_3B
+    layers:
+      layers.10: model.layers.10.post_attention_layernorm
+      layers.15: model.layers.15.post_attention_layernorm
+      layers.20: model.layers.20.post_attention_layernorm
+      norm: model.norm
+  qwen-2-5-omni-7b:
+    model: qwen-2-5-omni-7b
+    layers:
+      layers.5: model.layers.5.post_attention_layernorm
+      layers.10: model.layers.10.post_attention_layernorm
+      layers.15: model.layers.15.post_attention_layernorm
+      layers.20: model.layers.20.post_attention_layernorm
+      norm: model.norm
+  whisper:
+    model: whisper
+    layers:
+      layers.12: layers.12.fc2
+      layers.25: layers.25.fc2
+      layers.31: layers.31.fc2
+      norm: layer_norm
+  llama_3.2_1b:
+    model: Llama-3.2-1B
+    layers:
+      layers.7: model.layers.7
+      layers.11: model.layers.11
+      layers.15: model.layers.15
+  llama_3.2_3b:
+    model: Llama-3.2-3B
+    layers:
+      layers.7: model.layers.7
+      layers.11: model.layers.11
+      layers.15: model.layers.15
+      layers.19: model.layers.19
+      layers.23: model.layers.23
+  vjepa2:
+    model: vjepa2_avg_feat
+    layers:
+      layers.5: encoder.layer.5.norm1_avg
+      layers.15: encoder.layer.15.norm1_avg
+      layers.25: encoder.layer.25.norm1_avg
+      layers.35: encoder.layer.35.norm1_avg
+      norm: encoder.layernorm_avg
+include_features:
+- internvl3_14b/layers.30
+datasets:
+  train:
+    filter:
+      seasons:
+      - 1
+    sample_length: 64
+    num_samples: 2000
+    shuffle: true
+    seed: 42
+  val_s6:
+    filter:
+      seasons:
+      - 6
+      movies: []
+    sample_length: null
+    num_samples: null
+    shuffle: false
+  val_figures:
+    filter:
+      seasons: []
+      movies:
+      - figures
+    sample_length: null
+    num_samples: null
+    shuffle: false
+val_set_name: val_figures
+datasets_root: /workspace/data

output/two_stage_encoding/stage1_best.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98179fd2307efe54791f2b97ed5cf4efbb722aa292ab269c592e75a9b06513a5
+size 7833845

output/two_stage_encoding/stage2_epoch_0.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c4013c39fc311e3cb8ae3efb9745e96a68ecc9779a25e18885640e0d5e2ded13
+size 240515294

output/two_stage_encoding/stage2_epoch_5.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0477043b81bc838b582451d42a41c362e7ebaff7c5b4a8078e50720090701e9d
+size 240515294

output/two_stage_encoding/stage2_epoch_9.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a72612fcfce5d9b8866b66d9f5eb275d342422719c97f969ca7fcd510047b00
+size 240515294