Skip to content

[Bugfix][Model Runner V2] Preserve Mamba block table capacity under DCP #50287

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
jongukc wants to merge 2 commits into
base: main
Choose a base branch
from
Open

[Bugfix][Model Runner V2] Preserve Mamba block table capacity under DCP #50287

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
eb372b5
[Bugfix][Model Runner V2] Preserve Mamba block table capacity under DCP
jongukc Jul 27, 2026
3a91629
Merge branch 'main' into fix/v2-hybrid-dcp-block-table
jongukc Aug 5, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 8 additions & 1 deletion vllm/v1/worker/gpu/block_table.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -114,10 +114,17 @@ def append_block_ids(
start = self.num_blocks.np[i, req_index] if not overwrite else 0
block_ids = new_block_ids[i]
bpk = self.blocks_per_kv_block[i]
end = start + len(block_ids) * bpk
capacity = self.block_tables[i].gpu.shape[1]
if end > capacity:
raise RuntimeError(
f"Block table write for request {req_index}, group {i} "
f"exceeds row capacity ({end} > {capacity})"
)
if bpk > 1:
block_ids = [b * bpk + k for b in block_ids for k in range(bpk)]
self.block_tables[i].stage_write(req_index, start, block_ids)
self.num_blocks.np[i, req_index] = start + len(block_ids)
self.num_blocks.np[i, req_index] = end

def apply_staged_writes(self) -> None:
if self.num_kv_cache_groups == 1:
Expand Down
7 changes: 3 additions & 4 deletions vllm/v1/worker/gpu/model_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,7 +54,7 @@
from vllm.utils.mem_utils import DeviceMemoryProfiler, format_gib
from vllm.utils.torch_utils import STR_DTYPE_TO_TORCH_DTYPE
from vllm.v1.core.sched.output import GrammarOutput, SchedulerOutput
from vllm.v1.kv_cache_interface import KVCacheConfig, MambaSpec
from vllm.v1.kv_cache_interface import AttentionSpec, KVCacheConfig, MambaSpec
from vllm.v1.outputs import DraftTokenIds, ModelRunnerOutput
from vllm.v1.worker.cp_utils import check_attention_cp_compatibility
from vllm.v1.worker.gpu import pcp_manager as pcp
Expand Down Expand Up @@ -441,9 +441,8 @@ def initialize_kv_cache(self, kv_cache_config: KVCacheConfig) -> None:
# When using DCP, each request's KV cache is sharded among different ranks.
# As a result, one block on the current rank covers `block_size * cp_size`
# tokens in the full, global (unsharded) sequence.
max_num_blocks = cdiv(
block_table_max_model_len, spec.block_size * self.dcp_size
)
dcp_size = self.dcp_size if isinstance(spec, AttentionSpec) else 1
max_num_blocks = cdiv(block_table_max_model_len, spec.block_size * dcp_size)
# Align to a multiple of (128 / block_size) as required by some attention
# backends such as TRTLLM (#39324)
if spec.block_size <= 128:
Expand Down
Loading