Skip to content

[https://nvbugs/6368053][fix] (1) Broaden the assertion to is_eagle3() or is_eagle3_one_model() #15689

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

Closed
tensorrt-cicd wants to merge 1 commit into from
Closed

[https://nvbugs/6368053][fix] (1) Broaden the assertion to is_eagle3() or is_eagle3_one_model() #15689

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: 3 additions & 2 deletions tensorrt_llm/_torch/attention_backend/trtllm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1029,8 +1029,9 @@ def update_spec_dec_param(

# Case 2/3: static tree
elif self.is_spec_dec_tree and not self.is_spec_dec_dynamic_tree and spec_metadata is not None:
assert spec_metadata.spec_dec_mode.is_eagle3(
), "Tree decoding is only supported for Eagle3 now"
assert (spec_metadata.spec_dec_mode.is_eagle3()
or spec_metadata.spec_dec_mode.is_eagle3_one_model()
), "Tree decoding is only supported for Eagle3 now"

is_target_model = not getattr(spec_metadata, 'is_draft_model',
False)
Expand Down
6 changes: 6 additions & 0 deletions tensorrt_llm/_torch/pyexecutor/model_engine.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4929,6 +4929,12 @@ def forward(self,
is_spec_dec_mode = spec_metadata.spec_dec_mode.attention_need_spec_dec_mode(
spec_resource_manager, self.is_draft_model, self.attn_backend,
self.model_is_wrapped)
# Same tree-decoding sync as in _prepare_tp_inputs: the
# attention-warmup path calls forward() before _prepare_tp_inputs
# runs, so runtime_draft_len has not yet been raised to the total
# tree token count to match the warmup buffer sizing.
if not self.spec_config.is_linear_tree:
self.runtime_draft_len = self.max_total_draft_tokens
# Propagate runtime_draft_len (already set on self by py_executor)
# to spec_metadata so downstream code (eagle3, interface, trtllm) can read it.
spec_metadata.runtime_draft_len = self.runtime_draft_len
Expand Down
2 changes: 0 additions & 2 deletions tests/integration/test_lists/waives.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -368,8 +368,6 @@ perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-ltx2_blackwel
perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-ltx2_blackwell-ltx2_2stage_bf16_t2v_cfg2_ulysses4_compile_on] SKIP (https://nvbugs/6294413)
perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-ltx2_blackwell-ltx2_nvfp4_i2v_cfg2_ulysses4_compile_on] SKIP (https://nvbugs/6294413)
test_doc.py::test_url_validity SKIP (https://nvbugs/6215684)
test_e2e.py::test_draft_token_tree_quickstart_advanced_eagle3[Llama-3.1-8b-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct-EAGLE3-LLaMA3.1-Instruct-8B] SKIP (https://nvbugs/6368053)
test_e2e.py::test_draft_token_tree_quickstart_advanced_eagle3_depth_1_tree[Llama-3.1-8b-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct-EAGLE3-LLaMA3.1-Instruct-8B] SKIP (https://nvbugs/6368053)
test_e2e.py::test_multi_nodes_eval[DeepSeek-R1/DeepSeek-R1-0528-FP4-tp16-mmlu] SKIP (https://nvbugs/6276983)
test_e2e.py::test_multi_nodes_eval[Kimi-K2-Thinking-NVFP4-tp16-mmlu] SKIP (https://nvbugs/6276983)
test_e2e.py::test_multi_nodes_eval[MiniMax-M2-tp16-mmlu] SKIP (https://nvbugs/6373532)
Expand Down
Loading