.

.deepspeed_env

@@ -2,8 +2,8 @@ WANDB_BASE_URL=https://wandb.szaiai.com
 WANDB_API_KEY=local-701636f51b4741d3862007df5cf7f12cca53d8d1
 WANDB_PROJECT=ds-qwen3
 WANDB_ENTITY=hailin
-WANDB_GROUP=q3-32b-ds4-2025-09-05
-WANDB_NAME=q3-32b-lr2e-5-train3
+WANDB_GROUP=q3-32b-ds4-2025-09-24
+WANDB_NAME=q3-32b-lr2e-5-train1
 WANDB_RESUME=allow
 WANDB_INIT_TIMEOUT=300
 WANDB_DIR=/tmp/$USER/wandb

mm-zero3.sh

@@ -35,7 +35,7 @@ FORCE_COLOR=1 deepspeed --hostfile hostfile \
     --per_device_train_batch_size 1 \
     --gradient_accumulation_steps 1 \
     --learning_rate 2e-5 --weight_decay 0.1 --warmup_ratio 0.02 \
-    --max_steps 20 \
+    --max_steps 300 \
     --log_interval 1 \
     --gradient_checkpointing \
     --bf16 \
@@ -43,5 +43,10 @@ FORCE_COLOR=1 deepspeed --hostfile hostfile \
     --report_to wandb \
     --wandb_project ds-qwen3 \
     --eval_steps 10 \
+    --save_steps 10 \
+    --load_best_model_at_end \
+    --early_stopping_patience 5 \
+    --early_stopping_threshold 0.0 \  
+    --metric_for_best_model eval_loss \  
     --eval_data_glob "/home/test/datasets/my_corpus/test.jsonl"
 

train_sft_ds.py

@@ -28,6 +28,7 @@ from transformers import (
 from transformers.trainer_callback import TrainerCallback
 from transformers.trainer_utils import get_last_checkpoint
 from torch.optim import AdamW as TorchAdamW
+from transformers import EarlyStoppingCallback
 
 # ==== make sure CLI ninja/nvcc are reachable even in non-interactive ssh ====
 import site, shutil
@@ -444,6 +445,17 @@ def parse_args():
     ap.add_argument("--eval_steps", type=int, default=10,
                     help="Evaluate every N optimizer steps when eval_dataset is provided")    
     
+    ap.add_argument("--load_best_model_at_end", action="store_true",
+                    help="训练结束时自动加载最优 checkpoint")
+    ap.add_argument("--metric_for_best_model", type=str, default="eval_loss",
+                    help="用哪个指标选最优，默认 eval_loss")
+    ap.add_argument("--greater_is_better", action="store_true",
+                    help="是否指标越大越好；eval_loss 用 False（默认不传即可）")
+    ap.add_argument("--early_stopping_patience", type=int, default=0,
+                    help=">0 启用早停；单位是 eval 轮次数（非 step 数）")
+    ap.add_argument("--early_stopping_threshold", type=float, default=0.0,
+                    help="改进阈值，0 表示严格变好才算改进")    
+    
     return ap.parse_args()
 
 
@@ -921,7 +933,7 @@ def main():
         lr_scheduler_type="cosine",
         logging_steps=args.log_interval,
         save_steps=args.save_steps,
-        save_total_limit=2,
+        save_total_limit=None,
         deepspeed=(args.deepspeed if use_ds else None),
         dataloader_drop_last=False,
         dataloader_num_workers=0,
@@ -948,6 +960,21 @@ def main():
         "fp16": (torch.cuda.is_available() and not use_bf16),
     })
 
+    ta_sig = inspect.signature(TrainingArguments.__init__).parameters
+    if "save_strategy" in ta_sig:
+        ta_kwargs2["save_strategy"] = "steps"
+
+    ta_kwargs2.update({
+        "load_best_model_at_end": args.load_best_model_at_end,
+        "metric_for_best_model": args.metric_for_best_model,
+        "greater_is_better": args.greater_is_better,  # 对 eval_loss 保持 False（默认）
+        # "save_strategy": "steps",  # 与 eval_steps 对齐
+    })    
+
+
+    if args.early_stopping_patience > 0 and eval_dataset is None:
+        print("[warn] early_stopping_patience>0 但未提供 eval 数据集；早停将不会触发。", flush=True)
+
     training_args = TrainingArguments(**ta_kwargs2)    
 
     trainer_kwargs = {}
@@ -966,6 +993,13 @@ def main():
         **trainer_kwargs,
     )
 
+
+    if args.early_stopping_patience and args.early_stopping_patience > 0:
+        trainer.add_callback(EarlyStoppingCallback(
+            early_stopping_patience=args.early_stopping_patience,
+            early_stopping_threshold=args.early_stopping_threshold
+        ))
+
     trainer.add_callback(CsvLossLogger(csv_path=os.path.join(args.output_dir, "loss.csv")))