qwen config upgrade

evaluation/evaluation.py

@@ -88,8 +88,8 @@ def is_dataset_cached(dataset_name):
 
 
 def evaluate():
-    # model_name = "Qwen3-Embedding-0.6B"
-    model_name = "llama-embed-nemotron-8b"
+    model_name = "Qwen3-Embedding-0.6B"
+    # model_name = "llama-embed-nemotron-8b"
     # model_name = "embeddinggemma-300m"
     model = CustomModel(model_name)
 

train/qwen/a.sh

@@ -4,7 +4,7 @@
 nproc_per_node=1
 
 MLFLOW_TRACKING_URI=http://0.0.0.0:5004 \
-INFONCE_USE_BATCH=False \
+INFONCE_USE_BATCH=True \
 CUDA_VISIBLE_DEVICES=0 \
 NPROC_PER_NODE=$nproc_per_node \
 swift sft \
@@ -12,11 +12,12 @@ swift sft \
     --task_type embedding \
     --model_type qwen3_emb \
     --train_type lora \
-    --lora_rank 8 \
-    --lora_alpha 16 \
+    --lora_rank 16 \
+    --lora_alpha 32 \
     --target_modules all-linear \
-    --dataset my_local_dataset \
-    --custom_register_path $(pwd)/../../data/dataset/my_dataset_register.py \
+    --max_length 2048  \
+    --dataset v11_dataset_hn \
+    --custom_register_path $(pwd)/../../data/dataset/v11_dataset_hn/generated.py \
     --split_dataset_ratio 0.005 \
     --eval_strategy steps \
     --output_dir output \
@@ -27,7 +28,8 @@ swift sft \
     --per_device_train_batch_size 16 \
     --per_device_eval_batch_size 16 \
     --gradient_accumulation_steps 4 \
-    --learning_rate 2.4e-5 \
+    --learning_rate 3.0e-6 \
+    --lr_scheduler_type constant \
     --loss_type infonce \
     --label_names labels \
     --dataloader_drop_last true \

train/qwen/merge_model.py

@@ -4,6 +4,8 @@ import os
 from peft import PeftModel
 import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
+import shutil
+os.environ["PEFT_BOFT_FORCE_CPU"] = "1"
 
 def merge(base_model_path, peft_model_path, save_path):
     base_model = AutoModelForCausalLM.from_pretrained(base_model_path, torch_dtype="bfloat16")
@@ -11,14 +13,34 @@ def merge(base_model_path, peft_model_path, save_path):
     ft_model = ft_model.merge_and_unload()
     ft_model.save_pretrained(save_path)
 
+def copy_selected_items(src_path, dst_path, items):
+    os.makedirs(dst_path, exist_ok=True)
+
+    for item in items:
+        source_item = os.path.join(src_path, item)
+        dest_item = os.path.join(dst_path, item)
+
+        if not os.path.exists(source_item):
+            print(f"⚠ {item} در مسیر مبدا پیدا نشد!")
+            continue
+
+        if os.path.isdir(source_item):
+            shutil.copytree(source_item, dest_item, dirs_exist_ok=True)
+        elif os.path.isfile(source_item):
+            shutil.copy2(source_item, dest_item)
+
 def main():
     file_path = os.path.dirname(__file__)
 
     base_model_path = file_path + "/../../data/models/Qwen3-Embedding-0.6B/model"
-    peft_model_path = file_path + "/output/v1-20251122-184545/checkpoint-3434"
-    save_path = file_path + "/output/v1-20251122-184545/merged_checkpoint-3434"
+    peft_model_path = file_path + "/output/v17-20251202-223944/checkpoint-387"
+    save_path = file_path + "/output/v17-20251202-223944/merged_checkpoint-387"
     merge(base_model_path, peft_model_path, save_path)
 
+    items = ["1_Pooling", "config_sentence_transformers.json", "merges.txt", "modules.json", "README.md", "tokenizer_config.json", "tokenizer.json", 
+        "vocab.json"]
+    copy_selected_items(base_model_path, save_path, items)
+
 
 if __name__ == "__main__":
-    main()
+    main()