Training in progress, step 500

config.json

@@ -0,0 +1,34 @@
+{
+  "_name_or_path": "OFA-Sys/chinese-clip-vit-base-patch16",
+  "architectures": [
+    "ChineseCLIPImageDualModel"
+  ],
+  "initializer_factor": 1.0,
+  "initializer_range": 0.02,
+  "logit_scale_init_value": 2.6592,
+  "model_type": "chinese_clip",
+  "projection_dim": 512,
+  "text_config": {
+    "architectures": [
+      "ChineseCLIPTextModel"
+    ],
+    "bos_token_id": 0,
+    "directionality": "bidi",
+    "eos_token_id": 2,
+    "model_type": "chinese_clip_text_model",
+    "output_past": true,
+    "pooler_fc_size": 768,
+    "pooler_num_attention_heads": 12,
+    "pooler_num_fc_layers": 3,
+    "pooler_size_per_head": 128,
+    "pooler_type": "first_token_transform",
+    "vocab_size": 21128
+  },
+  "torch_dtype": "float32",
+  "transformers_version": "4.42.3",
+  "vision_config": {
+    "dropout": 0.0,
+    "model_type": "chinese_clip_vision_model",
+    "patch_size": 16
+  }
+}

config.txt

@@ -0,0 +1,19 @@
+python clip_aoi_train.py \
+    --output_dir ./aoi_clip_high_resolution_crossAttenttionFusion_fusin_gpt_random_sampler \
+    --model_name_or_path OFA-Sys/chinese-clip-vit-base-patch16\
+    --image_processor_name OFA-Sys/chinese-clip-vit-base-patch16 \
+    --tokenizer_name OFA-Sys/chinese-clip-vit-base-patch16 \
+    --train_file ./data_csv/AOI/train_high_resolution_gpt.csv \
+    --validation_file ./data_csv/AOI/valid_high_resolution_gpt.csv \
+    --image_column image_path --caption_column text --aoi_caption_column aoi_text\
+    --remove_unused_columns=False  --do_eval --do_train \
+    --per_device_train_batch_size=40 --per_device_eval_batch_size=20 \
+    --gradient_accumulation_steps=10 --overwrite_output_dir\
+    --num_train_epochs=200 \
+    --learning_rate=1e-5 --warmup_steps=0 --weight_decay 0.1 \
+    --save_total_limit 5  \
+    --eval_strategy steps --logging_steps 0.1 --report_to wandb --fp16  --push_to_hub --batch_eval_metrics \
+    --max_seq_length 153 --aoi_max_seq_length 384
+    #--model_name_or_path OFA-Sys/chinese-clip-vit-base-patch16
+    #--overwrite_output_dir  \
+    # aoi_clip_high_resolution_concate_fusin_gpt_random_sampler

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:536a190a6161a2aaf2741479c1dd845eead0e300cec45f1ab349f4c1a9b83f31
+size 1164561480

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a65c5cfb22fcd570dd5cbf7c58678cc789f306da61bd04f94281acec7b3825ea
+size 5240