clean log

Author: jiayyu

atom/model_engine/model_runner.py

@@ -1059,7 +1059,7 @@ def propose_draft_token_ids(
                 self.tokenID_processor.prev_token_ids = self._latest_tokens_gpu
                 next_token_ids = self._latest_tokens_gpu
             elif self._latest_tokens_gpu.ndim == 2: # spec
-                bs = len(batch.seqs)
+                bs = len(batch.req_ids)
                 col_indices = self._lasted_num_accept - 1
                 next_token_ids = self._latest_tokens_gpu[torch.arange(bs, device=self._latest_tokens_gpu.device), col_indices]
                 self.tokenID_processor.prev_token_ids = next_token_ids

atom/model_engine/scheduler.py

@@ -201,19 +201,15 @@ def postprocess(
             token_ids = prev_token_ids[seq.id]
             new_tokens = []
             if is_deferred_out:
-                print(f"{token_ids=}")
                 idx = seq.token_ids.index(self.eos_token_id)
                 seq.token_ids[idx:] = token_ids
-                print(f"{seq.token_ids=}")
 
                 if seq.output_tokens:
                     idx = seq.output_tokens.index(self.eos_token_id)
                     seq.output_tokens[idx:] = token_ids
-                    print(f"{seq.output_tokens=}")
 
                 else:
                     seq.output_tokens.extend(token_ids)
-                    print(f"{seq.output_tokens=}")
 
                 new_tokens = token_ids
             else:

atom/spec_decode/eagle.py

@@ -147,10 +147,6 @@ def propose(
         inputs_embeds = None
         input_ids = self.input_ids[:num_input_tokens]
 
-        # forwad
-        if str(input_ids.device) == "cuda:0":
-            print(f"draft model forward {input_ids=}")
-
         ret_hidden_states = self.model(
             input_ids=input_ids,
             positions=self.positions[:num_input_tokens],