Skip to content

[https://nvbugs/6891163][fix] Build embed_tokens as a full copy (no mapping) when enable_attention_dp and… #20089

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
trtllm-agent wants to merge 2 commits into
base: main
Choose a base branch
from
+18 −10
Open

[https://nvbugs/6891163][fix] Build embed_tokens as a full copy (no mapping) when enable_attention_dp and… #20089

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
464e9b0
[https://nvbugs/6891163][fix] Replicate Qwen3 embedding under attenti…
trtllm-agent Oct 10, 2026
fc65496
[nvbugs/6891163][chore] Remove stale waiver after fix
trtllm-agent Oct 28, 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
27 changes: 18 additions & 9 deletions tensorrt_llm/_torch/models/modeling_qwen3.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -208,15 +208,24 @@ def __init__(self,
config = self.model_config
self.mapping_with_cp = mapping_with_cp

self.embed_tokens = Embedding(
config.pretrained_config.vocab_size,
config.pretrained_config.hidden_size,
dtype=config.pretrained_config.torch_dtype,
mapping=config.mapping,
tensor_parallel_mode=TensorParallelMode.COLUMN,
gather_output=True,
allreduce_strategy=config.allreduce_strategy,
)
# Under attention DP each rank embeds its own tokens, so the embedding
# is replicated like the attention-DP lm_head it may be tied to.
if config.mapping.enable_attention_dp and not config.mapping.enable_lm_head_tp_in_adp:
self.embed_tokens = Embedding(
config.pretrained_config.vocab_size,
config.pretrained_config.hidden_size,
dtype=config.pretrained_config.torch_dtype,
)
else:
self.embed_tokens = Embedding(
config.pretrained_config.vocab_size,
config.pretrained_config.hidden_size,
dtype=config.pretrained_config.torch_dtype,
mapping=config.mapping,
tensor_parallel_mode=TensorParallelMode.COLUMN,
gather_output=True,
allreduce_strategy=config.allreduce_strategy,
)
self.layers = nn.ModuleList([
Qwen3DecoderLayer(
model_config,
Expand Down
1 change: 0 additions & 1 deletion tests/integration/test_lists/waives.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -192,7 +192,6 @@ unittest/executor/test_rpc_proxy.py SKIP (https://nvbugs/5605741)
unittest/executor/test_rpc_worker.py SKIP (https://nvbugs/5605741)
unittest/llmapi/test_llm_args.py::TestKvCacheManagerV2AutoResolution::test_required_v2_refuses_incompatible_disagg[nixl-cpp] SKIP (https://nvbugs/6917693)
unittest/llmapi/test_llm_args.py::TestKvCacheManagerV2AutoResolution::test_required_v2_refuses_incompatible_disagg[ucx-python] SKIP (https://nvbugs/6917693)
unittest/llmapi/test_llm_kv_cache_events.py::test_llm_api_attention_dp_kv_events SKIP (https://nvbugs/6891163)
unittest/llmapi/test_llm_pytorch.py::test_gqa_nemo_lora[None] SKIP (https://nvbugs/6162504)
unittest/llmapi/test_llm_pytorch.py::test_gqa_nemo_lora[cuda_graph_config0] SKIP (https://nvbugs/6162504)
unittest/llmapi/test_llm_pytorch.py::test_llm_disagg_streaming_gen_cancelled[PYTHON] SKIP (https://nvbugs/6903683)
Expand Down
Loading