fix: use numb_epoch key and correct model_prob doc

- Change num_epoch to numb_epoch in change_bias.py and train.py

- Fix model_prob default description in multi-task-training.md

Author: OutisLi

deepmd/tf/entrypoints/change_bias.py

@@ -195,7 +195,7 @@ def _change_bias_checkpoint_file(
     # Get stop_batch and origin_type_map like in train.py
     training_params = jdata.get("training", {})
     stop_batch = training_params.get("numb_steps")
-    num_epoch = training_params.get("num_epoch")
+    num_epoch = training_params.get("numb_epoch")
     if stop_batch is None and num_epoch is not None:
         if num_epoch <= 0:
             raise ValueError("training.num_epoch must be positive.")

deepmd/tf/entrypoints/train.py

@@ -259,7 +259,7 @@ def _do_work(
     # get training info
     training_params = jdata["training"]
     stop_batch = training_params.get("numb_steps")
-    num_epoch = training_params.get("num_epoch")
+    num_epoch = training_params.get("numb_epoch")
     if stop_batch is None:
         if num_epoch is None:
             raise ValueError(

doc/train/multi-task-training.md

@@ -79,7 +79,7 @@ Specifically, there are several parts that need to be modified:
 
 - (Optional) {ref}`training/model_prob <training/model_prob>`: The sampling weight settings corresponding to each `model_key`, i.e., the probability weight in the training step.
   You can specify any positive real number weight for each task. The higher the weight, the higher the probability of being sampled in each training.
-  This setting is optional, and if not set, tasks will be sampled with equal weights. It is only used when `num_epoch_dict` is not set.
+  This setting is optional, and if not set, tasks will be sampled with weights proportional to the number of systems in each task. It is only used when `num_epoch_dict` is not set.
 
 - (Optional) {ref}`training/num_epoch_dict <training/num_epoch_dict>`: The number of training epochs for each model branch, specified as a dictionary mapping `model_key` to epoch values (can be fractional).
   This allows different tasks to train for different numbers of epochs, which is particularly useful for multi-task fine-tuning scenarios

source/tests/pt/test_sampler.py

@@ -35,7 +35,7 @@ class _SerialPool:
     def __init__(self, *args, **kwargs) -> None:
         pass
 
-    def __enter__(self) -> "_SerialPool":
+    def __enter__(self) -> "_SerialPool":  # noqa: PYI034
         return self
 
     def __exit__(self, exc_type, exc, tb) -> bool: