biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Model][Spec Decode] Nemotron-H MTP and Mamba Speculative Decoding Support (#33726)

Browse Source 
Signed-off-by: Shahar Mor <smor@nvidia.com>
Signed-off-by: Benjamin Chislett <bchislett@nvidia.com>
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: Shahar Mor <smor@nvidia.com>
Co-authored-by: Roi Koren <roik@nvidia.com>
Co-authored-by: Lucas Wilkinson <lwilkins@redhat.com>
This commit is contained in:
Benjamin Chislett
2026-02-24 12:49:56 -05:00
committed by GitHub
parent a9e15e040d
commit f5972a872f
19 changed files with 799 additions and 157 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
tests/v1/attention/test_mamba_update_block_table.py
View File

@@ -41,6 +41,9 @@ def _make_vllm_config(block_size, max_model_len, max_num_seqs):
cudagraph_mode=CUDAGraphMode.FULL,
max_cudagraph_capture_size=None,
),
speculative_config=None,
num_speculative_tokens=0,
parallel_config=SimpleNamespace(decode_context_parallel_size=1),
scheduler_config=SimpleNamespace(max_num_seqs=max_num_seqs),
model_config=SimpleNamespace(max_model_len=max_model_len),
)
@@ -92,7 +95,10 @@ def test_update_block_table_copies_block_idx_to_persistent_buffers():
has_initial_states_p=None,
query_start_loc_p=None,
num_computed_tokens_p=None,
state_indices_tensor=builder_a.state_indices_tensor[:num_reqs],
state_indices_tensor_p=None,
query_start_loc_d=None,
num_accepted_tokens=None,
state_indices_tensor_d=builder_a.state_indices_tensor_d[:num_reqs],
block_idx_last_scheduled_token=(
builder_a.block_idx_last_scheduled_token[:num_reqs]
),