Skip to content

Bound DS4 W2 restaging and tier execution #5

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
OmarB97 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Bound DS4 W2 restaging and tier execution #5

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
3 changes: 3 additions & 0 deletions .gitattributes
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
# Unified diffs contain a one-character context marker on blank context lines.
# Validate whitespace after applying the patch to its source tree instead.
patch/*.patch -whitespace
5 changes: 3 additions & 2 deletions Dockerfile.sm120-v024
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,8 +72,9 @@ RUN pip install --no-cache-dir --break-system-packages flashinfer-python==0.6.14
--index-url https://flashinfer.ai/whl/cu130 \
"flashinfer-jit-cache==0.6.14+cu130"

# 3. our patch, applied to the installed package (pure-Python delta;
# includes the DSpark speculative-decoding backport)
# 3. Combined v0.24.0 patch: W2 streaming/recovery, the step-driven tier
# manager, the DSpark speculative backport, and SM120 fixes. One canonical
# patch keeps the image label an exact source identity.
COPY patch/vllm-moet-v0.24.0.patch /tmp/vllm-moet.patch
RUN cd "$(python3 -c 'import vllm, os; print(os.path.dirname(os.path.dirname(vllm.__file__)))')" \
&& git apply --verbose /tmp/vllm-moet.patch \
Expand Down
7 changes: 7 additions & 0 deletions patch/FILES.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,9 @@
# patch but not in the vllm fork branch the patch is generated from —
# merge it into the branch first; do not ship the regeneration.
csrc/nvfp4_ds_mla/concat_and_cache_nvfp4_ds_mla.cu
tests/model_executor/layers/quantization/test_moe_w2_cgroup_memory.py
tests/model_executor/layers/quantization/test_moe_w2_padded_routes.py
tests/model_executor/layers/quantization/test_moe_w2_step_pins.py
tools/nvfp4_flashinfer_sm120/README.md
tools/nvfp4_flashinfer_sm120/nvfp4_expand.cuh
tools/nvfp4_flashinfer_sm120/patch_flashinfer.py
Expand All @@ -12,6 +15,7 @@ vllm/compilation/cuda_graph.py
vllm/config/speculative.py
vllm/config/vllm.py
vllm/envs.py
vllm/forward_context.py
vllm/model_executor/layers/attention/mla_attention.py
vllm/model_executor/layers/quantization/fp8.py
vllm/model_executor/layers/quantization/modelopt.py
Expand All @@ -27,6 +31,9 @@ vllm/model_executor/layers/quantization/utils/moe_w2_store.py
vllm/model_executor/layers/quantization/utils/prefill_timers.py
vllm/model_executor/layers/quantization/utils/skinny_fp8_cubit.py
vllm/model_executor/layers/sparse_attn_indexer.py
vllm/model_executor/model_loader/__init__.py
vllm/model_executor/model_loader/default_loader.py
vllm/model_executor/model_loader/weight_utils.py
vllm/model_executor/models/deepseek_mtp.py
vllm/model_executor/models/qwen3_dflash.py
vllm/model_executor/models/qwen3_dspark.py
Expand Down
Loading