[Deprecation] Deprecate items scheduled for 0.29 (#55353)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
@@ -2377,8 +2377,7 @@ class GPUModelRunner(
|
||||
num_prompt_tokens_cpu = self.input_batch.num_prompt_tokens_cpu_tensor[
|
||||
:num_reqs_padded
|
||||
]
|
||||
seq_lens_cpu = self.optimistic_seq_lens_cpu[:num_reqs_padded]
|
||||
seq_lens_cpu_upper_bound = seq_lens_cpu
|
||||
seq_lens_cpu_upper_bound = self.optimistic_seq_lens_cpu[:num_reqs_padded]
|
||||
|
||||
# is_prefilling: True if request is still in prefill phase.
|
||||
# Used by mamba backends to distinguish actual decodes from
|
||||
@@ -2388,11 +2387,6 @@ class GPUModelRunner(
|
||||
# misclassify padding as prefill in CUDA graph mode.
|
||||
is_prefilling[num_reqs:] = False
|
||||
|
||||
if self.use_async_spec_decode:
|
||||
# GPU tensors are authoritative in async mode.
|
||||
seq_lens_cpu = None
|
||||
num_computed_tokens_cpu = None
|
||||
|
||||
# Compute mm_prefix bidirectional ranges before building
|
||||
# attention metadata so builders handle them during build().
|
||||
# By default, ranges exceeding sliding_window are skipped to prevent
|
||||
@@ -2466,8 +2460,6 @@ class GPUModelRunner(
|
||||
query_start_loc=self.query_start_loc.gpu[: num_reqs_padded + 1],
|
||||
query_start_loc_cpu=self.query_start_loc.cpu[: num_reqs_padded + 1],
|
||||
seq_lens=self.seq_lens[:num_reqs_padded],
|
||||
_seq_lens_cpu=seq_lens_cpu,
|
||||
_num_computed_tokens_cpu=num_computed_tokens_cpu,
|
||||
seq_lens_cpu_upper_bound=seq_lens_cpu_upper_bound,
|
||||
replayssm_decode_base_cpu=replayssm_decode_base_cpu,
|
||||
num_reqs=num_reqs_padded,
|
||||
|
||||
Reference in New Issue
Block a user