[chat] polish code note typo (#3612)

applications/Chat/coati/experience_maker/base.py

@@ -18,7 +18,7 @@ class Experience:
     action_log_probs: (B, A)
     values: (B)
     reward: (B)
-    advatanges: (B)
+    advantages: (B)
     attention_mask: (B, S)
     action_mask: (B, A)
 

applications/Chat/coati/models/lora.py

@@ -108,7 +108,7 @@ def convert_to_lora_recursively(module: nn.Module, lora_rank: int) -> None:
 
 class LoRAModule(nn.Module):
     """A LoRA module base class. All derived classes should call `convert_to_lora()` at the bottom of `__init__()`.
-    This calss will convert all torch.nn.Linear layer to LoraLinear layer.
+    This class will convert all torch.nn.Linear layer to LoraLinear layer.
 
     Args:
         lora_rank (int, optional): LoRA rank. 0 means LoRA is not applied. Defaults to 0.

applications/Chat/coati/ray/src/detached_trainer_ppo.py

@@ -29,7 +29,7 @@ class DetachedPPOTrainer(DetachedTrainer):
         lora_rank (int) : for actor / critic init
         train_batch_size (int, defaults to 8): the batch size to use for training
         train_batch_size (int, defaults to 8): the batch size to use for training
-        buffer_limit (int, defaults to 0): the max_size limitaiton of replay buffer
+        buffer_limit (int, defaults to 0): the max_size limitation of replay buffer
         buffer_cpu_offload (bool, defaults to True): whether to offload replay buffer to cpu
         eps_clip (float, defaults to 0.2): the clip coefficient of policy loss
         value_clip (float, defaults to 0.4): the clip coefficient of value loss

applications/Chat/coati/ray/src/experience_maker_holder.py

@@ -83,7 +83,7 @@ class ExperienceMakerHolder:
         chosen_trainer = None
         min_length = None
         if 'debug' in self.generate_kwargs and self.generate_kwargs['debug'] == True:
-            print("[maker] choosing tartget trainer")
+            print("[maker] choosing target trainer")
         while chosen_trainer is None:
             for target_trainer in self.target_trainer_list:
                 try:

applications/Chat/coati/replay_buffer/utils.py

@@ -15,7 +15,7 @@ class BufferItem:
     action_log_probs: (A)
     values: (1)
     reward: (1)
-    advatanges: (1)
+    advantages: (1)
     attention_mask: (S)
     action_mask: (A)
 

applications/Chat/coati/trainer/callbacks/performance_evaluator.py

@@ -114,7 +114,7 @@ class PerformanceEvaluator(Callback):
 
         # actor forward-backward, 3 means forward(1) + backward(2)
         self.learn_flop += self.actor_num_params * batch_size * seq_len * 2 * (3 + int(self.enable_grad_checkpoint))
-        # critic foward-backward
+        # critic forward-backward
         self.learn_flop += self.critic_num_params * batch_size * seq_len * 2 * (3 + int(self.enable_grad_checkpoint))
 
     def on_fit_end(self) -> None:

applications/Chat/coati/trainer/ppo.py

@@ -28,12 +28,12 @@ class PPOTrainer(Trainer):
         actor (Actor): the actor model in ppo algorithm
         critic (Critic): the critic model in ppo algorithm
         reward_model (nn.Module): the reward model in rlhf algorithm to make reward of sentences
-        initial_model (Actor): the initial model in rlhf algorithm to generate reference logits to limit the update of actor
+        initial_model (Actor): the initial model in rlhf algorithm to generate reference logics to limit the update of actor
         actor_optim (Optimizer): the optimizer to use for actor model
         critic_optim (Optimizer): the optimizer to use for critic model
         kl_coef (float, defaults to 0.1): the coefficient of kl divergence loss
         train_batch_size (int, defaults to 8): the batch size to use for training
-        buffer_limit (int, defaults to 0): the max_size limitaiton of replay buffer
+        buffer_limit (int, defaults to 0): the max_size limitation of replay buffer
         buffer_cpu_offload (bool, defaults to True): whether to offload replay buffer to cpu
         eps_clip (float, defaults to 0.2): the clip coefficient of policy loss
         vf_coef (float, defaults to 1.0): the coefficient of value loss
@@ -41,7 +41,7 @@ class PPOTrainer(Trainer):
         value_clip (float, defaults to 0.4): the clip coefficient of value loss
         experience_batch_size (int, defaults to 8): the batch size to use for experience generation
         max_epochs (int, defaults to 1): the number of epochs of training process
-        tokenier (Callable, optional): the tokenizer to use for tokenizing the input
+        tokenizer (Callable, optional): the tokenizer to use for tokenizing the input
         sample_replay_buffer (bool, defaults to False): whether to sample from replay buffer
         dataloader_pin_memory (bool, defaults to True): whether to pin memory for data loader
         callbacks (List[Callback], defaults to []): the callbacks to call during training process