CLIP Train update

CLIP_last_layer/checkpoints/epoch=0-step=68000.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:657fc27c3f97ae064e7850389cc5f28876a1e584b771eebceed259bf12e1cf01
+size 17576378479

CLIP_last_layer/config.json

@@ -0,0 +1 @@
+{"lm_model_name": "meta-llama/Meta-Llama-3-8B-Instruct", "lr": 0.0001, "weight_decay": 0.01, "warmup_steps": 100, "vision_encoder_name": "clip", "vision_hidden_size": 1024, "freeze_vision": true, "use_lora": true, "vision_layer": -1, "num_workers": 8, "val_num_workers": 1, "shuffle_buffer": 10000, "resampled": false, "max_length": 32, "training_epochs": 1, "batch_size": 16, "devices": 2, "lora_r": 8, "lora_alpha": 16, "lora_dropout": 0.05, "num_sanity_val_steps": 2, "val_check_interval": 1000, "limit_val_batches": 16, "num_training_step": 68750}

CLIP_last_layer/events.out.tfevents.1764619008.ubuntu

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db912c8932c42e3b40f212ebd97a1366b93a363c30f367424bd9f62648ace140
+size 1678295

CLIP_last_layer/hparams.yaml

@@ -0,0 +1,24 @@
+config:
+  lm_model_name: meta-llama/Meta-Llama-3-8B-Instruct
+  lr: 0.0001
+  weight_decay: 0.01
+  warmup_steps: 100
+  vision_hidden_size: 1024
+  freeze_vision: true
+  use_lora: true
+  vision_layer: -1
+  num_workers: 8
+  val_num_workers: 1
+  shuffle_buffer: 10000
+  resampled: false
+  max_length: 32
+  training_epochs: 1
+  batch_size: 16
+  devices: 2
+  lora_r: 8
+  lora_alpha: 16
+  lora_dropout: 0.05
+  num_sanity_val_steps: 2
+  val_check_interval: 1000
+  limit_val_batches: 16
+  num_training_step: 68750

version_0_vggt_last_layer/config.json

@@ -1 +0,0 @@
-{"lm_model_name": "meta-llama/Meta-Llama-3-8B-Instruct", "lr": 0.0001, "weight_decay": 0.01, "warmup_steps": 100, "vision_hidden_size": 2048, "freeze_vision": true, "use_lora": true, "vision_layer": -1, "num_workers": 8, "val_num_workers": 1, "shuffle_buffer": 10000, "resampled": false, "max_length": 32, "training_epochs": 1, "batch_size": 16, "devices": 2, "lora_r": 8, "lora_alpha": 16, "lora_dropout": 0.05, "num_sanity_val_steps": 2, "val_check_interval": 1000, "limit_val_batches": 16, "num_training_step": 68750}

vggt_last_layer/config.json

@@ -0,0 +1 @@
+{"lm_model_name": "meta-llama/Meta-Llama-3-8B-Instruct", "lr": 0.0001, "weight_decay": 0.01, "warmup_steps": 100, "vision_encoder_name": "vggt", "vision_hidden_size": 2048, "freeze_vision": true, "use_lora": true, "vision_layer": -1, "num_workers": 8, "val_num_workers": 1, "shuffle_buffer": 10000, "resampled": false, "max_length": 32, "training_epochs": 1, "batch_size": 16, "devices": 2, "lora_r": 8, "lora_alpha": 16, "lora_dropout": 0.05, "num_sanity_val_steps": 2, "val_check_interval": 1000, "limit_val_batches": 16, "num_training_step": 68750}