Skip to content

test: Waive non-Llama Eagle tests - #3309

Merged
kaiyux merged 1 commit into
NVIDIA:mainfrom
brb-nv:user/brb/waive-eagle-tests
Apr 7, 2025
Merged

test: Waive non-Llama Eagle tests#3309
kaiyux merged 1 commit into
NVIDIA:mainfrom
brb-nv:user/brb/waive-eagle-tests

Conversation

@brb-nv

@brb-nv brb-nv commented Apr 6, 2025

Copy link
Copy Markdown
Collaborator

This MR waives Eagle tests for non-Llama architectures. Currently, Eagle implementations are specific to Llama.
https://github.com/NVIDIA/TensorRT-LLM/blob/main/tensorrt_llm/models/eagle/model.py#L578

@brb-nv
brb-nv force-pushed the user/brb/waive-eagle-tests branch 2 times, most recently from 1dfc024 to 6b20454 Compare April 6, 2025 20:50
@brb-nv

brb-nv commented Apr 6, 2025

Copy link
Copy Markdown
Collaborator Author

/bot run

@brb-nv
brb-nv requested a review from kaiyux April 6, 2025 22:52
@tensorrt-cicd

tensorrt-cicd commented Apr 6, 2025

Copy link
Copy Markdown
Collaborator

PR_Github #1239 [ run ] triggered by Bot

@tensorrt-cicd

tensorrt-cicd commented Apr 7, 2025

Copy link
Copy Markdown
Collaborator

PR_Github #1239 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #932 completed with status: 'SUCCESS'

@brb-nv @kaiyux
test: Waive non-Llama Eagle tests
7347b4a 
Signed-off-by: Balaram Buddharaju <169953907+brb-nv@users.noreply.github.com>
@kaiyux
kaiyux force-pushed the user/brb/waive-eagle-tests branch from 6b20454 to 7347b4a Compare April 7, 2025 01:13
@kaiyux
kaiyux enabled auto-merge (squash) April 7, 2025 01:14
@kaiyux

kaiyux commented Apr 7, 2025

Copy link
Copy Markdown
Member

/bot reuse-pipeline

@tensorrt-cicd

tensorrt-cicd commented Apr 7, 2025

Copy link
Copy Markdown
Collaborator

PR_Github #1246 [ reuse-pipeline ] triggered by Bot

@tensorrt-cicd

tensorrt-cicd commented Apr 7, 2025

Copy link
Copy Markdown
Collaborator

PR_Github #1246 [ reuse-pipeline ] completed with state SUCCESS
Reusing PR_Github #1239 for commit 7347b4a

@kaiyux
kaiyux merged commit 017361c into NVIDIA:main Apr 7, 2025
sarattha pushed a commit to sarattha/TensorRT-LLM that referenced this pull request Apr 9, 2025
@brb-nv @sarattha
test: Waive non-Llama Eagle tests (NVIDIA#3309)
09e95d9 
Signed-off-by: sarattha <j.sarattha@gmail.com>
wu1du2 pushed a commit to wu1du2/TensorRT-LLM that referenced this pull request May 11, 2025
@brb-nv
test: Waive non-Llama Eagle tests (NVIDIA#3309)
4f24892
haow-nv added a commit to haow-nv/TensorRT-LLM that referenced this pull request Jun 29, 2026
@haow-nv
[None][feat] CuteDSL MLA decode: persistent + split-KV alignment (fla…
0158e3d 
…shinfer NVIDIA#2743/NVIDIA#3309)

- is_persistent = not is_var_seq: variable-seq paged decode now uses
  non-persistent tile scheduling instead of the hardcoded persistent path.
- split-KV: fold-aware simplified heuristic (sq_eff via compute_fold_sq_ratio),
  CUDA-graph-safe split selection.
- mla_helpers: non-persistent work-tile coordinate fix (b/s divmod + divisor),
  previously dormant because the path was never taken under hardcoded persistent.
- moe_scheduler: attention-DP MoE-dispatch all-gather deadlock fix -- apply the
  empty-chunk substitution consistently across all ranks so the variable-size
  all-gather 'sizes' stay identical on every rank.
- tests: cute_dsl_mla_decode updated (incl. fold_sq / small-batch split_kv).

Signed-off-by: haow <haow@nvidia.com>
haow-nv added a commit to haow-nv/TensorRT-LLM that referenced this pull request Jul 6, 2026
@haow-nv
[None][feat] CuteDSL MLA decode: persistent + split-KV alignment (fla…
9f678b4 
…shinfer NVIDIA#2743/NVIDIA#3309)

- is_persistent = not is_var_seq: variable-seq paged decode now uses
  non-persistent tile scheduling instead of the hardcoded persistent path.
- split-KV: fold-aware simplified heuristic (sq_eff via compute_fold_sq_ratio),
  CUDA-graph-safe split selection.
- mla_helpers: non-persistent work-tile coordinate fix (b/s divmod + divisor),
  previously dormant because the path was never taken under hardcoded persistent.
- moe_scheduler: attention-DP MoE-dispatch all-gather deadlock fix -- apply the
  empty-chunk substitution consistently across all ranks so the variable-size
  all-gather 'sizes' stay identical on every rank.
- tests: cute_dsl_mla_decode updated (incl. fold_sq / small-batch split_kv).

Signed-off-by: haow <haow@nvidia.com>
haow-nv added a commit to haow-nv/TensorRT-LLM that referenced this pull request Jul 8, 2026
@haow-nv
[None][feat] CuteDSL MLA decode: persistent + split-KV alignment (fla…
af7006b 
…shinfer NVIDIA#2743/NVIDIA#3309)

- is_persistent = not is_var_seq: variable-seq paged decode now uses
  non-persistent tile scheduling instead of the hardcoded persistent path.
- split-KV: fold-aware simplified heuristic (sq_eff via compute_fold_sq_ratio),
  CUDA-graph-safe split selection.
- mla_helpers: non-persistent work-tile coordinate fix (b/s divmod + divisor),
  previously dormant because the path was never taken under hardcoded persistent.
- moe_scheduler: attention-DP MoE-dispatch all-gather deadlock fix -- apply the
  empty-chunk substitution consistently across all ranks so the variable-size
  all-gather 'sizes' stay identical on every rank.
- tests: cute_dsl_mla_decode updated (incl. fold_sq / small-batch split_kv).

Signed-off-by: haow <haow@nvidia.com>
haow-nv added a commit to haow-nv/TensorRT-LLM that referenced this pull request Jul 14, 2026
@haow-nv
[None][feat] CuteDSL MLA decode: persistent + split-KV alignment (fla…
9c7bf07 
…shinfer NVIDIA#2743/NVIDIA#3309)

- is_persistent = not is_var_seq: variable-seq paged decode now uses
  non-persistent tile scheduling instead of the hardcoded persistent path.
- split-KV: fold-aware simplified heuristic (sq_eff via compute_fold_sq_ratio),
  CUDA-graph-safe split selection.
- mla_helpers: non-persistent work-tile coordinate fix (b/s divmod + divisor),
  previously dormant because the path was never taken under hardcoded persistent.
- moe_scheduler: attention-DP MoE-dispatch all-gather deadlock fix -- apply the
  empty-chunk substitution consistently across all ranks so the variable-size
  all-gather 'sizes' stay identical on every rank.
- tests: cute_dsl_mla_decode updated (incl. fold_sq / small-batch split_kv).

Signed-off-by: haow <haow@nvidia.com>
haow-nv added a commit to haow-nv/TensorRT-LLM that referenced this pull request Jul 21, 2026
@haow-nv
[None][feat] CuteDSL MLA decode: persistent + split-KV alignment (fla…
6a0269d 
…shinfer NVIDIA#2743/NVIDIA#3309)

- is_persistent = not is_var_seq: variable-seq paged decode now uses
  non-persistent tile scheduling instead of the hardcoded persistent path.
- split-KV: fold-aware simplified heuristic (sq_eff via compute_fold_sq_ratio),
  CUDA-graph-safe split selection.
- mla_helpers: non-persistent work-tile coordinate fix (b/s divmod + divisor),
  previously dormant because the path was never taken under hardcoded persistent.
- moe_scheduler: attention-DP MoE-dispatch all-gather deadlock fix -- apply the
  empty-chunk substitution consistently across all ranks so the variable-size
  all-gather 'sizes' stay identical on every rank.
- tests: cute_dsl_mla_decode updated (incl. fold_sq / small-batch split_kv).

Signed-off-by: haow <haow@nvidia.com>
haow-nv added a commit to haow-nv/TensorRT-LLM that referenced this pull request Jul 23, 2026
@haow-nv
[None][feat] CuteDSL MLA decode: persistent + split-KV alignment (fla…
6748399 
…shinfer NVIDIA#2743/NVIDIA#3309)

- is_persistent = not is_var_seq: variable-seq paged decode now uses
  non-persistent tile scheduling instead of the hardcoded persistent path.
- split-KV: fold-aware simplified heuristic (sq_eff via compute_fold_sq_ratio),
  CUDA-graph-safe split selection.
- mla_helpers: non-persistent work-tile coordinate fix (b/s divmod + divisor),
  previously dormant because the path was never taken under hardcoded persistent.
- moe_scheduler: attention-DP MoE-dispatch all-gather deadlock fix -- apply the
  empty-chunk substitution consistently across all ranks so the variable-size
  all-gather 'sizes' stay identical on every rank.
- tests: cute_dsl_mla_decode updated (incl. fold_sq / small-batch split_kv).

Signed-off-by: haow <haow@nvidia.com>
haow-nv added a commit to haow-nv/TensorRT-LLM that referenced this pull request Jul 29, 2026
@haow-nv
[None][feat] CuteDSL MLA decode: persistent + split-KV alignment (fla…
99a748b 
…shinfer NVIDIA#2743/NVIDIA#3309)

- is_persistent = not is_var_seq: variable-seq paged decode now uses
  non-persistent tile scheduling instead of the hardcoded persistent path.
- split-KV: fold-aware simplified heuristic (sq_eff via compute_fold_sq_ratio),
  CUDA-graph-safe split selection.
- mla_helpers: non-persistent work-tile coordinate fix (b/s divmod + divisor),
  previously dormant because the path was never taken under hardcoded persistent.
- moe_scheduler: attention-DP MoE-dispatch all-gather deadlock fix -- apply the
  empty-chunk substitution consistently across all ranks so the variable-size
  all-gather 'sizes' stay identical on every rank.
- tests: cute_dsl_mla_decode updated (incl. fold_sq / small-batch split_kv).

Signed-off-by: haow <haow@nvidia.com>
haow-nv added a commit to haow-nv/TensorRT-LLM that referenced this pull request Aug 3, 2026
@haow-nv
[None][feat] CuteDSL MLA decode: persistent + split-KV alignment (fla…
286c7ed 
…shinfer NVIDIA#2743/NVIDIA#3309)

- is_persistent = not is_var_seq: variable-seq paged decode now uses
  non-persistent tile scheduling instead of the hardcoded persistent path.
- split-KV: fold-aware simplified heuristic (sq_eff via compute_fold_sq_ratio),
  CUDA-graph-safe split selection.
- mla_helpers: non-persistent work-tile coordinate fix (b/s divmod + divisor),
  previously dormant because the path was never taken under hardcoded persistent.
- moe_scheduler: attention-DP MoE-dispatch all-gather deadlock fix -- apply the
  empty-chunk substitution consistently across all ranks so the variable-size
  all-gather 'sizes' stay identical on every rank.
- tests: cute_dsl_mla_decode updated (incl. fold_sq / small-batch split_kv).

Signed-off-by: haow <haow@nvidia.com>
haow-nv added a commit to haow-nv/TensorRT-LLM that referenced this pull request Aug 4, 2026
@haow-nv
[None][feat] CuteDSL MLA decode: persistent + split-KV alignment (fla…
0fe7413 
…shinfer NVIDIA#2743/NVIDIA#3309)

- is_persistent = not is_var_seq: variable-seq paged decode now uses
  non-persistent tile scheduling instead of the hardcoded persistent path.
- split-KV: fold-aware simplified heuristic (sq_eff via compute_fold_sq_ratio),
  CUDA-graph-safe split selection.
- mla_helpers: non-persistent work-tile coordinate fix (b/s divmod + divisor),
  previously dormant because the path was never taken under hardcoded persistent.
- moe_scheduler: attention-DP MoE-dispatch all-gather deadlock fix -- apply the
  empty-chunk substitution consistently across all ranks so the variable-size
  all-gather 'sizes' stay identical on every rank.
- tests: cute_dsl_mla_decode updated (incl. fold_sq / small-batch split_kv).

Signed-off-by: haow <haow@nvidia.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@kaiyux kaiyux kaiyux approved these changes

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@brb-nv @tensorrt-cicd @kaiyux