Skip to content

[AutoDeploy] more robust handling of attention interface and input nodes - #31

Merged
lucaslie merged 2 commits into
feat/staging_05_22from
ll/attn_if_updated
May 23, 2025
Merged

[AutoDeploy] more robust handling of attention interface and input nodes#31
lucaslie merged 2 commits into
feat/staging_05_22from
ll/attn_if_updated

Conversation

@lucaslie

@lucaslie lucaslie commented May 23, 2025

Copy link
Copy Markdown

supersedes and closes #4369 (was already reviewed and approved)

lucaslie added 2 commits May 22, 2025 20:30
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
@lucaslie
lucaslie force-pushed the ll/attn_if_updated branch from 952a715 to 8875b5a Compare May 23, 2025 03:45
@lucaslie
lucaslie merged commit 122f4a2 into feat/staging_05_22 May 23, 2025
@lucaslie
lucaslie deleted the ll/attn_if_updated branch May 23, 2025 03:54
lucaslie added a commit that referenced this pull request May 23, 2025
…33)

* Add Phi3.5-mini patch and revert gemm_fusion change (rebased) (#28)

* add phi3.5 patch

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* revert gemm fusion change

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* update README

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

---------

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* [AutoDeploy] correct random init (#30)

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

* [AutoDeploy] more robust handling of attention interface and input nodes (#31)

* [AutoDeploy] more robust handling of attention interface and input nodes

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

* updated description for AttentionInterface

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

---------

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

* Several Bug Fix: bias missing for linear/lift canonicalize_graph to meta device/enhance attention op args extraction (#27)

* add bias patch for torch.ops.linear.simple

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* lift canonicalize_graph with shape prop to meta device

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* fix attention custom op default args/kwargs extract

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* add rope use case of the util file

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

---------

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* Fix for attention argument dropout_p='None'  (#29)

* updated attention args logic tp handle case in which default values for dropout_p is not captured by the graph

Signed-off-by: Suguna Velury <178320438+sugunav14@users.noreply.github.com>

* Formatting fix

Signed-off-by: Suguna Velury <178320438+sugunav14@users.noreply.github.com>

* Updated args assignment logic in match_causal_attn_mask

Signed-off-by: Suguna Velury <178320438+sugunav14@users.noreply.github.com>

---------

Signed-off-by: Suguna Velury <178320438+sugunav14@users.noreply.github.com>

* [AutoDeploy] Llama-4 support + AutoModelForImageTextToText (#32)

* [AutoDeploy] Llama-4 support

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

* WAR: Rope Matcher Issue

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

---------

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

* Apply suggestions from code review

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

---------

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
Signed-off-by: Suguna Velury <178320438+sugunav14@users.noreply.github.com>
Co-authored-by: Fridah-nv <201670829+Fridah-nv@users.noreply.github.com>
Co-authored-by: sugunav14 <178320438+sugunav14@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
lucaslie added a commit that referenced this pull request May 27, 2025
…33)

* Add Phi3.5-mini patch and revert gemm_fusion change (rebased) (#28)

* add phi3.5 patch

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* revert gemm fusion change

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* update README

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

---------

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* [AutoDeploy] correct random init (#30)

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

* [AutoDeploy] more robust handling of attention interface and input nodes (#31)

* [AutoDeploy] more robust handling of attention interface and input nodes

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

* updated description for AttentionInterface

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

---------

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

* Several Bug Fix: bias missing for linear/lift canonicalize_graph to meta device/enhance attention op args extraction (#27)

* add bias patch for torch.ops.linear.simple

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* lift canonicalize_graph with shape prop to meta device

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* fix attention custom op default args/kwargs extract

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* add rope use case of the util file

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

---------

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* Fix for attention argument dropout_p='None'  (#29)

* updated attention args logic tp handle case in which default values for dropout_p is not captured by the graph

Signed-off-by: Suguna Velury <178320438+sugunav14@users.noreply.github.com>

* Formatting fix

Signed-off-by: Suguna Velury <178320438+sugunav14@users.noreply.github.com>

* Updated args assignment logic in match_causal_attn_mask

Signed-off-by: Suguna Velury <178320438+sugunav14@users.noreply.github.com>

---------

Signed-off-by: Suguna Velury <178320438+sugunav14@users.noreply.github.com>

* [AutoDeploy] Llama-4 support + AutoModelForImageTextToText (#32)

* [AutoDeploy] Llama-4 support

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

* WAR: Rope Matcher Issue

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

---------

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

* Apply suggestions from code review

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

---------

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
Signed-off-by: Suguna Velury <178320438+sugunav14@users.noreply.github.com>
Co-authored-by: Fridah-nv <201670829+Fridah-nv@users.noreply.github.com>
Co-authored-by: sugunav14 <178320438+sugunav14@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant