opt ppo model

README.md

@@ -83,7 +83,7 @@ python train_ppo.py  --load_sft_model "./out_sft/rwkv-190.pth" --load_rm_model "
 --ctx_len 1024 --epoch_steps 200 --epoch_count 1000 --epoch_begin 0 --epoch_save 2 \
 --micro_bsz 2 --n_layer 24 --n_embd 2048 --pre_ffn 0 --head_qk 0 \
 --lr_init 1e-5 --lr_final 1e-5 --warmup_steps 0 --beta1 0.9 --beta2 0.999 --adam_eps 1e-8 \
---accelerator gpu --devices 1 --precision bf16 --strategy deepspeed_stage_2_offload --grad_cp 0 \
+--accelerator gpu --devices 1 --precision bf16 --strategy deepspeed_stage_2_offload --grad_cp 1 \
 --my_qa_mask 1
 ```
 

src/rlhf/ppo.py

@@ -71,8 +71,7 @@ class ActorCritic(nn.Module):
         state,
         max_seq_len,
         eos_token = None,
-        return_values = False,
-        **kwargs
+        return_values = False
     ):
         # 产生一条 response，相当于采取了一次 action
         actions = self.actor.generate(
@@ -324,10 +323,8 @@ class RLHF(pl.LightningModule):
     def generate(
         self,
         max_seq_len,
-        *args,
         prompt,
-        num_samples = 4,  # sample 4 per prompt and select the one with highest reward
-        **kwargs
+        num_samples = 4  # sample 4 per prompt and select the one with highest reward
     ):
         ''' 未参与训练，仅推理时使用
         '''
@@ -345,10 +342,8 @@ class RLHF(pl.LightningModule):
             _
         ) = self.actor_critic.generate(
             prompt,
-            *args,
             max_seq_len = max_seq_len,
-            return_values = False,
-            **kwargs
+            return_values = False
         )
 
         rewards = self.reward_model(