Create training_config.yml

training_config.yml

@@ -0,0 +1,78 @@
+config:
+  (): colpali_engine.trainer.colmodel_training.ColModelTrainingConfig
+  output_dir: !path ../../../models/biqwen2_5_1ep_colpali_train_5neg_128bs_7b_2e4
+  processor:
+    (): colpali_engine.utils.transformers_wrappers.AllPurposeWrapper
+    class_to_instanciate: !ext colpali_engine.models.BiQwen2_5_Processor
+    pretrained_model_name_or_path:  "Qwen/Qwen2.5-VL-7B-Instruct" # "./models/paligemma-3b-mix-448"
+    # num_image_tokens: 2048
+    # max_length: 50
+  num_negatives: 5 
+
+  model:
+    (): colpali_engine.utils.transformers_wrappers.AllPurposeWrapper
+    class_to_instanciate: !ext colpali_engine.models.BiQwen2_5
+    pretrained_model_name_or_path: "Qwen/Qwen2.5-VL-7B-Instruct"
+    torch_dtype:  !ext torch.bfloat16
+    use_cache: false
+    attn_implementation: "flash_attention_2"
+#    device_map: "auto"
+#    quantization_config:
+#      (): transformers.BitsAndBytesConfig
+#      load_in_4bit: true
+#      bnb_4bit_quant_type: "nf4"
+#      bnb_4bit_compute_dtype:  "bfloat16"
+#      bnb_4bit_use_double_quant: true
+
+  dataset_loading_func: !ext colpali_engine.utils.dataset_transformation.load_train_set_ir_negs
+  eval_dataset_loader: !import ../data/test_data.yaml
+
+  # max_length: 50
+  run_eval: true
+  loss_func:
+    (): colpali_engine.loss.bi_encoder_losses.BiEncoderLoss
+  tr_args:
+    (): transformers.training_args.TrainingArguments
+    output_dir: null
+    overwrite_output_dir: true
+    num_train_epochs: 1 
+    # this is global batch size 
+    per_device_train_batch_size: 128 
+    gradient_checkpointing: true
+    gradient_checkpointing_kwargs: { "use_reentrant": false }
+    ddp_find_unused_parameters: false
+    resume_from_checkpoint: false
+    # gradient_checkpointing: true
+    # 6 x 8 gpus = 48 batch size
+    # gradient_accumulation_steps: 4
+    per_device_eval_batch_size: 16
+    eval_strategy: "no"
+    dataloader_num_workers: 0
+    log_level: "info"
+    accelerator_config:
+      split_batches: true
+    # bf16: true
+    save_steps: 500
+    logging_steps: 10
+    eval_steps: 100
+    warmup_ratio: 0.01
+    learning_rate: 2e-4
+    save_total_limit: 1
+    # resume_from_checkpoint: true
+    optim: "paged_adamw_8bit"
+    # fsdp: "shard_grad_op"
+    # wandb logging
+    run_name: "biqwen2_5_1ep_colpali_train_5neg_128bs_7b_2e4"
+    report_to: "wandb"
+
+
+  peft_config:
+    (): peft.LoraConfig
+    r: 32
+    lora_alpha: 32
+    lora_dropout: 0.1
+    init_lora_weights: "gaussian"
+    bias: "none"
+    task_type: "FEATURE_EXTRACTION"
+    target_modules: '(.*(model).*(down_proj|gate_proj|up_proj|k_proj|q_proj|v_proj|o_proj).*$|.*(custom_text_proj).*$)'
+    # target_modules: '(.*(language_model).*(down_proj|gate_proj|up_proj|k_proj|q_proj|v_proj|o_proj).*$|.*(custom_text_proj).*$)'