opt ppo model

src/model.py

@@ -521,7 +521,7 @@ class RWKV(pl.LightningModule):
         sample_num_times = max(1, seq_len - prompt.shape[-1])
 
         for _ in tqdm(range(sample_num_times), desc="gen responses"):
-            pad_idx = torch.tensor([[eos_token] * (self.args.n_embd - out.shape[-1])])
+            pad_idx = torch.tensor([[eos_token] * (self.args.ctx_len - out.shape[-1])])
             query_idx = torch.cat((out, pad_idx), dim=-1)
             logits, embeds = self.forward(query_idx, ppo_train=True)
             logits, embeds = logits[:, -1], embeds[:, -1]