[Metax][CI] e2e ci tests enable cuda graph (#6401)

fastdeploy/model_executor/pre_and_post_process.py

@@ -911,6 +911,7 @@ def rebuild_padding(
             seq_lens_decoder,
             seq_lens_encoder,
             batch_id_per_token_output,
+            cu_seqlens_q_output,
             first_token_out,
             enable_logprob,
         )

fastdeploy/worker/metax_model_runner.py

@@ -1264,8 +1264,8 @@ class MetaxModelRunner(ModelRunnerBase):
             batch_id_per_token,
             cu_seqlens_q,
             cu_seqlens_k,
-            output_cum_offsets,
+            cu_seqlens_q_output,
-            output_padding_offset,
+            batch_id_per_token_output,
         ) = pre_process(
             token_num_cpu,
             self.share_inputs["input_ids"],
@@ -1284,8 +1284,8 @@ class MetaxModelRunner(ModelRunnerBase):
 
         # For speculative decoding
         if self.speculative_decoding:
-            self.share_inputs["output_cum_offsets"].copy_(output_cum_offsets, False)
+            self.share_inputs["cu_seqlens_q_output"].copy_(cu_seqlens_q_output, False)
-            self.share_inputs["output_padding_offset"].copy_(output_padding_offset, False)
+            self.share_inputs["batch_id_per_token_output"].copy_(batch_id_per_token_output, False)
 
         # Initialize forward meta data
         self.initialize_forward_meta(is_dummy_or_profile_run=is_dummy_or_profile_run)
@@ -1896,10 +1896,8 @@ class MetaxModelRunner(ModelRunnerBase):
                     self.share_inputs["seq_lens_this_time"],
                     self.share_inputs["seq_lens_decoder"],
                     self.share_inputs["seq_lens_encoder"],
-                    (
+                    (self.share_inputs["batch_id_per_token_output"] if self.speculative_decoding else None),
-                        self.share_inputs["output_padding_offset"] if self.speculative_decoding else None
+                    (self.share_inputs["cu_seqlens_q_output"] if self.speculative_decoding else None),
-                    ),  # speculative decoding requires
-                    self.model_config.max_model_len,
                 )
                 self._dummy_sampler_run(hidden_states, model_output, accept_all_drafts, reject_all_drafts)
 
@@ -2337,8 +2335,8 @@ class MetaxModelRunner(ModelRunnerBase):
                 self.share_inputs["seq_lens_this_time"],
                 self.share_inputs["seq_lens_decoder"],
                 self.share_inputs["seq_lens_encoder"],
-                (self.share_inputs["output_padding_offset"] if self.speculative_decoding else None),
+                (self.share_inputs["batch_id_per_token_output"] if self.speculative_decoding else None),
-                self.model_config.max_model_len,
+                (self.share_inputs["cu_seqlens_q_output"] if self.speculative_decoding else None),
             )
 
             # 4. Compute logits, Sample

tests/metax_ci/test_ernie_21b.py

@@ -31,7 +31,7 @@ class TestErnie21B(unittest.TestCase):
             load_choices="default_v1",
             # enable_prefix_caching=False,
             disable_custom_all_reduce=True,
-            graph_optimization_config={"use_cudagraph": False, "graph_opt_level": 0},
+            # graph_optimization_config={"use_cudagraph": False, "graph_opt_level": 0},
         )
 
         cls.sampling_params = fastdeploy.SamplingParams(top_p=0.95, max_tokens=256, temperature=0.6)

tests/metax_ci/test_ernie_28b_vl.py

@@ -37,7 +37,7 @@ class TestErnie28BVL(unittest.TestCase):
             quantization="wint8",
             disable_custom_all_reduce=True,
             # enable_prefix_caching=False,
-            graph_optimization_config={"use_cudagraph": False, "graph_opt_level": 0},
+            # graph_optimization_config={"use_cudagraph": False, "graph_opt_level": 0},
             limit_mm_per_prompt={"image": 100},
             reasoning_parser="ernie-45-vl",
             load_choices="default_v1",