graph_trainer: support MinimalAsyncEP graph chunking - #3956
Draft
sanketpurandare wants to merge 4 commits into
Draft
graph_trainer: support MinimalAsyncEP graph chunking#3956sanketpurandare wants to merge 4 commits into
sanketpurandare wants to merge 4 commits into
Conversation
[ghstack-poisoned]
This was referenced Jul 21, 2026
meta-cla
Bot
added
the
CLA Signed
[ghstack-poisoned]
…aph chunking" ## Summary Extend the generic FX chunking path to MinimalAsyncEP. Discover chunk roots through graph-visible token-exchange launches, then assign MinimalAsyncEP buffer sets with a separate schema-driven post-chunk pass. The generic graph cloning algorithm remains backend-independent. Eager and graph chunking feed the same ownership and launch/wait metadata to the shared EP scheduler. ## Why Buffer assignment is a backend storage concern, not part of generic region cloning. Keeping it in a small post-pass preserves the graph chunking contract and avoids MinimalAsyncEP special cases in the large cloning pass. ## Test Plan - Validate token-exchange count invariants for the generalized scheduler. - Run an exact two-rank DeepSeek-v3 comparison between graph and eager MinimalAsyncEP chunking. - Compare both loss and gradient norm for 20 steps with forced-balanced routing and regional Inductor. [ghstack-poisoned]
This was referenced Jul 29, 2026
sanketpurandare
added a commit
that referenced
this pull request
Aug 1, 2026
## Summary Extend the generic FX chunking path to MinimalAsyncEP. Discover chunk roots through graph-visible token-exchange launches, then assign MinimalAsyncEP buffer sets with a separate schema-driven post-chunk pass. The generic graph cloning algorithm remains backend-independent. Eager and graph chunking feed the same ownership and launch/wait metadata to the shared EP scheduler. ## Why Buffer assignment is a backend storage concern, not part of generic region cloning. Keeping it in a small post-pass preserves the graph chunking contract and avoids MinimalAsyncEP special cases in the large cloning pass. ## Test Plan - Validate token-exchange count invariants for the generalized scheduler. - Run an exact two-rank DeepSeek-v3 comparison between graph and eager MinimalAsyncEP chunking. - Compare both loss and gradient norm for 20 steps with forced-balanced routing and regional Inductor. ghstack-source-id: 6ecba47 Pull Request resolved: #3956
sanketpurandare
added a commit
that referenced
this pull request
Aug 1, 2026
## Summary Extend the generic FX chunking path to MinimalAsyncEP. Discover chunk roots through graph-visible token-exchange launches, then assign MinimalAsyncEP buffer sets with a separate schema-driven post-chunk pass. The generic graph cloning algorithm remains backend-independent. Eager and graph chunking feed the same ownership and launch/wait metadata to the shared EP scheduler. ## Why Buffer assignment is a backend storage concern, not part of generic region cloning. Keeping it in a small post-pass preserves the graph chunking contract and avoids MinimalAsyncEP special cases in the large cloning pass. ## Test Plan - Validate token-exchange count invariants for the generalized scheduler. - Run an exact two-rank DeepSeek-v3 comparison between graph and eager MinimalAsyncEP chunking. - Compare both loss and gradient norm for 20 steps with forced-balanced routing and regional Inductor. ghstack-source-id: 6ecba47 Pull Request resolved: #3956
sanketpurandare
added a commit
that referenced
this pull request
Aug 1, 2026
## Summary Extend the generic FX chunking path to MinimalAsyncEP. Discover chunk roots through graph-visible token-exchange launches, then assign MinimalAsyncEP buffer sets with a separate schema-driven post-chunk pass. The generic graph cloning algorithm remains backend-independent. Eager and graph chunking feed the same ownership and launch/wait metadata to the shared EP scheduler. ## Why Buffer assignment is a backend storage concern, not part of generic region cloning. Keeping it in a small post-pass preserves the graph chunking contract and avoids MinimalAsyncEP special cases in the large cloning pass. ## Test Plan - Validate token-exchange count invariants for the generalized scheduler. - Run an exact two-rank DeepSeek-v3 comparison between graph and eager MinimalAsyncEP chunking. - Compare both loss and gradient norm for 20 steps with forced-balanced routing and regional Inductor. ghstack-source-id: 6ecba47 Pull Request resolved: #3956
sanketpurandare
added a commit
that referenced
this pull request
Aug 1, 2026
## Summary Extend the generic FX chunking path to MinimalAsyncEP. Discover chunk roots through graph-visible token-exchange launches, then assign MinimalAsyncEP buffer sets with a separate schema-driven post-chunk pass. The generic graph cloning algorithm remains backend-independent. Eager and graph chunking feed the same ownership and launch/wait metadata to the shared EP scheduler. ## Why Buffer assignment is a backend storage concern, not part of generic region cloning. Keeping it in a small post-pass preserves the graph chunking contract and avoids MinimalAsyncEP special cases in the large cloning pass. ## Test Plan - Validate token-exchange count invariants for the generalized scheduler. - Run an exact two-rank DeepSeek-v3 comparison between graph and eager MinimalAsyncEP chunking. - Compare both loss and gradient norm for 20 steps with forced-balanced routing and regional Inductor. ghstack-source-id: 6ecba47 Pull Request resolved: #3956
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stack from ghstack (oldest at bottom):
Summary
Extend the generic FX chunking path to MinimalAsyncEP. Discover chunk roots
through graph-visible token-exchange launches, then assign MinimalAsyncEP
buffer sets with a separate schema-driven post-chunk pass.
The generic graph cloning algorithm remains backend-independent. Eager and
graph chunking feed the same ownership and launch/wait metadata to the shared
EP scheduler.
Why
Buffer assignment is a backend storage concern, not part of generic region
cloning. Keeping it in a small post-pass preserves the graph chunking contract
and avoids MinimalAsyncEP special cases in the large cloning pass.
Test Plan
MinimalAsyncEP chunking.
and regional Inductor.