Skip to content

Navigation Menu

Appearance settings

- AI CODE CREATION
  GitHub CopilotWrite better code with AI
  GitHub Copilot appDirect agents from issue to merge
  MCP Registry^NewIntegrate external tools
- DEVELOPER WORKFLOWS
  ActionsAutomate any workflow
  CodespacesInstant dev environments
  IssuesPlan and track work
  Code ReviewManage code changes
- APPLICATION SECURITY
  GitHub Advanced SecurityFind and fix vulnerabilities
  Code securitySecure your code as you build
  Secret protectionStop leaks before they start
- EXPLORE
  Why GitHub
  Documentation
  Blog
  Changelog
  Marketplace
View all features
- BY COMPANY SIZE
  Enterprises
  Small and medium teams
  Startups
  Nonprofits
- BY USE CASE
  App Modernization
  DevSecOps
  DevOps
  CI/CD
  View all use cases
- BY INDUSTRY
  Healthcare
  Financial services
  Manufacturing
  Government
  View all industries
View all solutions
- EXPLORE BY TOPIC
  AI
  Software Development
  DevOps
  Security
  View all topics
- EXPLORE BY TYPE
  Customer stories
  Events & webinars
  Ebooks & reports
  Business insights
  GitHub Skills
- SUPPORT & SERVICES
  Documentation
  Customer support
  Community forum
  Trust center
  Partners
View all resources
- COMMUNITY
  GitHub SponsorsFund open source developers
- PROGRAMS
  Security Lab
  Maintainer Community
  Accelerator
  GitHub Stars
  Archive Program
- REPOSITORIES
  Topics
  Trending
  Collections
- ENTERPRISE SOLUTIONS
  Enterprise platformAI-powered developer platform
- AVAILABLE ADD-ONS
  GitHub Advanced SecurityEnterprise-grade security features
  Copilot for BusinessEnterprise-grade AI features
  Premium SupportEnterprise-grade 24/7 support
Pricing

Search code, repositories, users, issues, pull requests...

Search

Clear

Search syntax tips

Provide feedback

We read every piece of feedback, and take your input very seriously.

Include my email address so I can be contacted

Saved searches

Use saved searches to filter your results more quickly

Name

Query

To see all available qualifiers, see our documentation.

Appearance settings

You signed in with another tab or window. Reload to refresh your session. You signed out in another tab or window. Reload to refresh your session. You switched accounts on another tab or window. Reload to refresh your session.

Dismiss alert

Uh oh!

There was an error while loading. Please reload this page.

OpenPipe / ART Public

Notifications You must be signed in to change notification settings
Fork 911
Star 10.2k

Code
Issues 62
Pull requests 59
Actions
Projects
Security and quality
Insights

Additional navigation options

Code
Issues
Pull requests
Actions
Projects
Security and quality
Insights

feat: add TrainerRank#731

Closed

bradhilton wants to merge 115 commits into

mainOpenPipe/ART:mainfrom

feat/trainer-rank-gdn-treeOpenPipe/ART:feat/trainer-rank-gdn-treeCopy head branch name to clipboard

Conversation Commits115 (115)Checks Files changed

Closed

feat: add TrainerRank#731
bradhilton wants to merge 115 commits into
mainfrom
feat/trainer-rank-gdn-tree

Commits

Commits on Jun 22, 2026

fix: support non-monotonic CP block masks
bradhilton
committed
fix: remove uncommitted packing dependency from CP tests
bradhilton
committed
feat: add TrainerRank and generic tree GDN
bradhilton
committed
fix: satisfy quality checks
bradhilton
committed
fix: balance tree GDN CP plans
bradhilton
committed
feat: add dynamic LoRA slots for TrainerRank
bradhilton
committed
fix: resolve vllm runtime python for codec tests
bradhilton
committed
style: apply TrainerRank formatting
bradhilton
committed
fix: guard dynamic slot optimizer reductions
bradhilton
committed
Merge remote-tracking branch 'origin/main' into feat/trainer-rank-gdn-tree
bradhilton
committed

Commits on Jun 23, 2026

test: add megatron review readiness harness
bradhilton
committed
test: add Austin-focused TrainerRank validation
bradhilton
committed
test: harden Austin validation harness
bradhilton
committed
fix: prune false exact CP mask blocks
bradhilton
committed
fix: refine full exact CP mask blocks
bradhilton
committed
perf: avoid unnecessary exact CP mask refinement
bradhilton
committed
fix: keep exact CP refinement for deep trees
bradhilton
committed
test: make CP mask parity slice aware
bradhilton
committed
fix: broadcast slice-aware mask parity
bradhilton
committed
perf: vectorize depth-one CP mask refinement
bradhilton
committed
perf: keep depth-one CP mask refinement vectorized
bradhilton
committed
fix: handle empty CP correctness shards
bradhilton
committed
fix: collect perf metadata on all ranks
bradhilton
committed
bench: use CP stage masks for flex timing
bradhilton
committed
fix: benchmark production sparse flex path
bradhilton
committed
fix: mirror production flex stage padding in review bench
bradhilton
committed
fix: default flex review bench to qwen head dim
bradhilton
committed
perf: skip exact refinement for homogeneous mask blocks
bradhilton
committed
perf: use generic interval block mask
bradhilton
committed
feat: add adaptive trainer rank microbatches
bradhilton
committed
fix: handle uninitialized trainer rank topology
bradhilton
committed
fix: report trainer rank oom call context
bradhilton
committed

Commits on Jun 24, 2026

bench: add TrainerRank adaptive microbatch perf cases
bradhilton
committed
fix: keep masked TrainerRank outputs graph-connected
bradhilton
committed
bench: expose TrainerRank adaptive memory knobs
bradhilton
committed
bench: profile TrainerRank adaptive microbatches
bradhilton
committed
bench: split adaptive selector profile
bradhilton
committed
perf: cache adaptive TrainerRank plans
bradhilton
committed
perf: preflight adaptive TrainerRank candidate plans
bradhilton
committed
perf: speed up adaptive packed-token estimates
bradhilton
committed
perf: defer adaptive TrainerRank plan materialization
bradhilton
committed
fix: restore quality checks
bradhilton
committed
test: add TrainerRank weird-shape fast gate
bradhilton
committed
fix: format TrainerRank fast gate
bradhilton
committed
refactor: extract TrainerRank adaptive planner
bradhilton
committed
refactor: trim TrainerRank planning surface
bradhilton
committed
fix: preserve masked target logprobs for shared rows
bradhilton
committed
refactor: collapse TrainerRank adaptive cache key
bradhilton
committed
refactor: collapse TrainerRank memory profile checks
bradhilton
committed
fix: update TrainerRank adaptive perf profiler
bradhilton
committed
refactor: collapse TrainerRank memory checks
bradhilton
committed
refactor: trim TrainerRank forwarding checks
bradhilton
committed
refactor: simplify TrainerRank memory signature
bradhilton
committed
refactor: keep target oracle helper in dev
bradhilton
committed
refactor: localize topology oracle logits helper
bradhilton
committed
refactor: use hidden head path for target logprobs
bradhilton
committed
refactor: trim TrainerRank head and planner paths
bradhilton
committed
refactor: return full TrainerRank head outputs
bradhilton
committed
refactor: simplify TrainerRank slot grad plumbing
bradhilton
committed
refactor: cache TrainerRank checkpoint slot params
bradhilton
committed
refactor: collapse LoRA parallel layout builders
bradhilton
committed
refactor: unify shared-prefix planning
bradhilton
committed
refactor: collapse GDN bucket builders
bradhilton
committed
refactor: remove stale context parallel helpers
bradhilton
committed
refactor: unify context parallel peer exchange
bradhilton
committed
refactor: centralize adapter export traversal
bradhilton
committed
refactor: use generic context parallel pair planning
bradhilton
committed
refactor: remove unused context planner metadata
bradhilton
committed

Commits on Jun 25, 2026

chore: fix review perf harness
bradhilton
committed
refactor: inline trainer rank microbatch planner
bradhilton
committed
perf: prune shared-prefix block-mask refinement
bradhilton
committed
refactor: simplify trainer rank planning records
bradhilton
committed
perf: skip full-slice minmax in block masks
bradhilton
committed
refactor: simplify context parallel state records
bradhilton
committed
refactor: simplify gdn tree state lookup
bradhilton
committed
perf: stabilize adaptive trainer windows
bradhilton
committed
perf: keep adaptive window size stable across tails
bradhilton
committed
perf: balance adaptive trainer windows
bradhilton
committed
chore: trace adaptive trainer perf windows
bradhilton
committed
chore: trace adaptive train-step windows
bradhilton
committed
chore: cap review block-mask validation
bradhilton
committed
refactor: simplify trainer rank head helpers
bradhilton
committed
perf: bound adaptive memory profile growth
bradhilton
committed
perf: speed shared-prefix common-prefix scan
bradhilton
committed
chore: add review perf thresholds
bradhilton
committed
refactor: simplify block-mask interval refinement
bradhilton
committed
refactor: unify trainer rank local stats kernels
bradhilton
committed
refactor: simplify lora slot support
bradhilton
committed
refactor: simplify shared-expert lora wrapper
bradhilton
committed
refactor: collapse lora moe wrappers
bradhilton
committed
refactor: simplify trainer rank lora plumbing
bradhilton
committed
refactor: collapse context parallel dispatch glue
bradhilton
committed
refactor: trim context parallel runtime plans
bradhilton
committed
refactor: unify trainer rank head projection
bradhilton
committed
refactor: simplify trainer rank planning
bradhilton
committed
refactor: simplify block mask validation
bradhilton
committed
refactor: remove alternate cp planner strategies
bradhilton
committed
refactor: trim trainer rank gdn surfaces
bradhilton
committed
refactor: shrink shared prefix parser metadata
bradhilton
committed
refactor: use shared prefix pack in trainer rank
bradhilton
committed
refactor: trim trainer rank packing helpers
bradhilton
committed
refactor: share trainer rank grad sync helpers
bradhilton
committed
refactor: unify gdn tree bucket planning
bradhilton
committed
refactor: trim adaptive planner cache keys
bradhilton
committed
refactor: inline stale trainer rank helpers
bradhilton
committed
test: allow tensor fields in gdn layout fixture
bradhilton
committed
fix: type triton topk launch constants
bradhilton
committed
test: handle fused expert fc1 lora in oracle
bradhilton
committed
test: canonicalize padded forward trace rows
bradhilton
committed
fix: run cp tree child buckets recurrently
bradhilton
committed
Revert "fix: run cp tree child buckets recurrently"
bradhilton
committed
fix: keep unchained gdn subtrees colocated
bradhilton
committed
fix: type triton topk launches
bradhilton
committed
fix: type grouped lora calls
bradhilton
committed
fix: scope trainer rank memory checks to model group
bradhilton
committed

Footer

© 2026 GitHub, Inc.

Footer navigation

Terms
Privacy
Security
Status
Community
Docs
Contact

You can’t perform that action at this time.