Skip to content

[None][fix] Fix DeepSeekV32 test_fp8_blockscale[baseline_mtp1] OOM on Blackwell #12823

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
sunnyqgg wants to merge 1 commit into
base: main
Choose a base branch
from
+6 −1
Open

[None][fix] Fix DeepSeekV32 test_fp8_blockscale[baseline_mtp1] OOM on Blackwell #12823

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
5 changes: 4 additions & 1 deletion tests/integration/defs/accuracy/test_llm_api_pytorch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3030,7 +3030,10 @@ def test_fp8_blockscale(self, tp_size, pp_size, ep_size, mtp_nextn, fp8kv,
if get_sm_version() == 100 or get_sm_version() == 103:
moe_backend = "DEEPGEMM" if moe_backend == "_DEFAULT" else moe_backend
moe_config = MoeConfig(backend=moe_backend, max_num_tokens=16384)
kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.6)
# MTP layers consume extra memory; reduce KV cache fraction to
# avoid OOM during forward pass on Blackwell (NVBug 5955792).
fraction = 0.55 if mtp_nextn > 0 else 0.6
kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=fraction)
else:
if moe_backend != "_DEFAULT":
pytest.skip("Not supported MoE backend!")
Expand Down
2 changes: 2 additions & 0 deletions tests/integration/test_lists/waives.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -273,6 +273,8 @@ full:RTXPro6000D/accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp
accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False] SKIP (https://nvbugs/5955765)
accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_no_kv_cache_reuse[quant_dtype=none-mtp_nextn=2-fp8kv=False-attention_dp=True-cuda_graph=True-overlap_scheduler=True] SKIP (https://nvbugs/5955773)
accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_fp8_blockscale[baseline_mtp1] SKIP (https://nvbugs/5955792)
accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False] SKIP (https://nvbugs/5961736)
unittest/auto_deploy/multigpu/transformations/library/test_tp_sharding.py::test_sharding[Linear-torch_dist_all_gather-True-False-2] SKIP (https://nvbugs/5961739)
full:RTXPro6000D/accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=True] SKIP (https://nvbugs/5961814)
examples/test_visual_gen.py::test_visual_gen_quickstart SKIP (https://nvbugs/5963896)
accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_2_model_mtp SKIP (https://nvbugs/5966585)
Expand Down
Loading