Skip to content

Navigation Menu

Appearance settings

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.

modal-projects / foundry Public

forked from foundry-org/foundry

Notifications You must be signed in to change notification settings
Fork 0
Star 0

Code
Pull requests 7
Actions
Projects
Security and quality
Insights

Additional navigation options

Code
Pull requests
Actions
Projects
Security and quality
Insights

[tp] Consolidate branches and validate vLLM/SGLang - #6

#6

Draft

rchalamala wants to merge 168 commits into

mainmodal-projects/foundry:mainfrom

cursor/tp-upstream-consolidation-fe33modal-projects/foundry:cursor/tp-upstream-consolidation-fe33Copy head branch name to clipboard

Conversation Commits168 (168)Checks Files changed

Draft

[tp] Consolidate branches and validate vLLM/SGLang#6
rchalamala wants to merge 168 commits into
mainfrom
cursor/tp-upstream-consolidation-fe33

Commits

Commits on Jul 23, 2026

docs: design TP upstream consolidation

cursoragent
and
rchalamala
committed
docs: plan TP upstream consolidation

cursoragent
and
rchalamala
committed
[fix] add ep ipc fix, deepep works without nvl_bytes=0

xenshinu
authored and
cursoragent
committed
[sglang] add DeepEP no-fabric IPC policy

devin-ai-integration[bot]
authored and
cursoragent
committed
[sglang] add no-fabric DeepEP IPC recipe

devin-ai-integration[bot]
authored and
cursoragent
committed
[sglang] fail clearly on old DeepEP API

devin-ai-integration[bot]
authored and
cursoragent
committed
[sglang] move inspect import to module top

devin-ai-integration[bot]
authored and
cursoragent
committed
[sglang] enforce DeepEP use_fabric requirement at Buffer construction

devin-ai-integration[bot]
authored and
cursoragent
committed
[sglang] add experimental dense tensor-parallel recipe

devin-ai-integration[bot]
authored and
cursoragent
committed
[fix] fix workspace mismatch when using fa2

xenshinu
authored and
cursoragent
committed
test: add reproducible SGLang TP validation

cursoragent
and
rchalamala
committed
fix: keep TP NCCL transport symmetric

cursoragent
and
rchalamala
committed
fix: pin TP validation to immutable revisions

cursoragent
and
rchalamala
committed
fix: align TP graph mode across phases

cursoragent
and
rchalamala
committed
docs: mark SGLang tensor parallel validated

cursoragent
and
rchalamala
committed
test: strengthen tensor parallel verification

cursoragent
and
rchalamala
committed
fix: harden graph and IPC restoration

cursoragent
and
rchalamala
committed
fix: version preallocated IPC chunks

cursoragent
and
rchalamala
committed
test: require complete TP replay evidence

cursoragent
and
rchalamala
committed
fix: pass default graph pool explicitly

cursoragent
and
rchalamala
committed
test: compare semantic TP graph archives

cursoragent
and
rchalamala
committed
docs: make TP validation invocation portable

cursoragent
and
rchalamala
committed
test: unpack loaded graph results

cursoragent
and
rchalamala
committed
fix: pin TP validation seed

cursoragent
and
rchalamala
committed
docs: explain deterministic TP validation

cursoragent
and
rchalamala
committed
test: compare deterministic TP graph structure

cursoragent
and
rchalamala
committed
docs: align TP scope with upstream roadmap

cursoragent
and
rchalamala
committed
fix: preserve Python CUDAGraph public API

cursoragent
and
rchalamala
committed
test: require exact RNG replay restoration

cursoragent
and
rchalamala
committed
fix: make experimental recipe defaults portable

cursoragent
and
rchalamala
committed
fix: make DeepEP matrix harness portable

cursoragent
and
rchalamala
committed
test: add reproducible vLLM TP validation

cursoragent
and
rchalamala
committed
style: format vLLM TP harness

cursoragent
and
rchalamala
committed
fix: pin compatible vLLM CUDA wheel

cursoragent
and
rchalamala
committed
fix: load torch before vLLM extension check

cursoragent
and
rchalamala
committed
fix: verify vLLM extension without CUDA driver

cursoragent
and
rchalamala
committed
test: cover zero-alignment VMM reservations

cursoragent
and
rchalamala
committed
fix: preserve VMM cursor for default alignment

cursoragent
and
rchalamala
committed
fix: keep vLLM TP graphs on NCCL collectives

cursoragent
and
rchalamala
committed
fix: pass vLLM TP compilation config atomically

cursoragent
and
rchalamala
committed
test: use stable vLLM TP acceptance signals

cursoragent
and
rchalamala
committed
test: bound vLLM TP output comparison

cursoragent
and
rchalamala
committed
fix: normalize default VMM reserve alignment

cursoragent
and
rchalamala
committed
test: harden vLLM TP validation evidence

cursoragent
and
rchalamala
committed
style: format TP validation fixes

cursoragent
and
rchalamala
committed
fix: correct concurrent TP evidence type

cursoragent
and
rchalamala
committed
[sglang] port graph-capture hook to init_forward_metadata 3-method ABC (#26735)

devin-ai-integration[bot]
authored and
cursoragent
committed
style: keep SGLang compatibility imports scoped

cursoragent
and
rchalamala
committed
test: cover SGLang attention metadata compatibility

cursoragent
and
rchalamala
committed
test: verify SGLang metadata allocation symmetry

cursoragent
and
rchalamala
committed
fix: enforce validated TP communication profile

cursoragent
and
rchalamala
committed
fix: reject surplus TP recipe arguments

cursoragent
and
rchalamala
committed
docs: define validated tensor parallel scope

cursoragent
and
rchalamala
committed
docs: clarify tensor parallel evidence provenance

cursoragent
and
rchalamala
committed
chore: satisfy repository lint gates

cursoragent
and
rchalamala
committed
test: add reproducible native CUDA validation

cursoragent
and
rchalamala
committed
test: harden native validation preflight

cursoragent
and
rchalamala
committed
docs: add SGLang current-main port design and plan

cursoragent
and
rchalamala
committed
docs: pin SGLang model revision in port plan

cursoragent
and
rchalamala
committed
docs: number SGLang main port tasks

cursoragent
and
rchalamala
committed
feat(sglang): activate via plugin entry point and pin upstream main

cursoragent
and
rchalamala
committed
fix(sglang): keep inactive plugin imports lightweight

cursoragent
and
rchalamala
committed
feat(sglang): register main profile and rank hooks

cursoragent
and
rchalamala
committed
feat(sglang): reuse saved memory pool config on load

cursoragent
and
rchalamala
committed
fix(sglang): serialize shared warmup state atomically

cursoragent
and
rchalamala
committed
fix(sglang): validate saved memory pool values

cursoragent
and
rchalamala
committed
fix(sglang): reject incompatible warmup state

cursoragent
and
rchalamala
committed
feat(sglang): bind SAVE/LOAD to FullCudaGraphBackend and ShapeKey

cursoragent
and
rchalamala
committed
fix(sglang): fail closed on incomplete ShapeKey archive keys

cursoragent
and
rchalamala
committed
fix(sglang): interleave per-shape wrapper alloc with per-index graph load

cursoragent
and
rchalamala
committed
fix(graph): propagate background build failures at per-entry finish

cursoragent
and
rchalamala
committed
test(sglang): assert clean-checkout target resolution

cursoragent
and
rchalamala
committed
fix(sglang): normalize classmethod target identity

cursoragent
and
rchalamala
committed
fix(sglang): make Modal target contract self-contained

cursoragent
and
rchalamala
committed
fix(sglang): provide CUDA driver to Modal contract

cursoragent
and
rchalamala
committed
test(sglang): keep inactive Modal check CPU-only

cursoragent
and
rchalamala
committed
docs(sglang): clarify Modal contract resources

cursoragent
and
rchalamala
committed
test(sglang): harden TP harness for pinned-main plugin architecture

cursoragent
and
rchalamala
committed
docs(sglang): document zero-edit plugin activation and pinned-main scope

cursoragent
and
rchalamala
committed
test(sglang): prove archive immutability, group teardown, and exact TP evidence

cursoragent
and
rchalamala
committed
docs(sglang): correct pinned-main scope, harness pin, and hook descriptions

cursoragent
and
rchalamala
committed
test(sglang): byte-level archive immutability and generic native-error scan

cursoragent
and
rchalamala
committed
fix(modal): mount sibling helper modules into SGLang/vLLM TP images

cursoragent
and
rchalamala
committed
fix(modal): mount sibling helpers into cleanup_run's slim image

cursoragent
and
rchalamala
committed
fix(sglang): two-window fatal scan to end SIGINT-teardown false positives

cursoragent
and
rchalamala
committed
test(sglang): flag mid-run subprocess crashes pre-teardown only

cursoragent
and
rchalamala
committed
test(sglang): readiness-aware Python-failure scanning

cursoragent
and
rchalamala
committed
test(sglang): recognize recovered retry blocks explicitly

cursoragent
and
rchalamala
committed
docs(sglang): correct stale HARD_FATAL_PATTERN comment

cursoragent
and
rchalamala
committed
docs(sglang): plan symmetric TP restore treatment

cursoragent
and
rchalamala
committed
fix(sglang): restore graphs with stable symmetric TP collectives

cursoragent
and
rchalamala
committed
fix(sglang): harden TP profile shape guard and symmetric gather eligibility

cursoragent
and
rchalamala
committed
test(sglang): fix gather return-value storage assertion

cursoragent
and
rchalamala
committed
test(sglang): add focused symmetric TP restore treatment

cursoragent
and
rchalamala
committed
fix(sglang): harden focused TP treatment evidence

cursoragent
and
rchalamala
committed
fix(sglang): classify focused startup failures

cursoragent
and
rchalamala
committed
fix(sglang): preserve focused startup causes

cursoragent
and
rchalamala
committed
fix(sglang): classify nested startup exceptions

cursoragent
and
rchalamala
committed
fix(sglang): close traceback at log boundaries

cursoragent
and
rchalamala
committed
fix(sglang): recognize CR-prefixed markers

cursoragent
and
rchalamala
committed
fix(sglang): verify archived symmetric collectives

cursoragent
and
rchalamala
committed
fix(sglang): drive LOAD through the pinned upstream capture lifecycle

cursoragent
and
rchalamala
committed
fix(sglang): stop gating LOAD acceptance on native capture progress; NONE-transparent cleanup

cursoragent
and
rchalamala
committed
fix(sglang): make cleanup part of verification

cursoragent
and
rchalamala
committed
fix(sglang): fail closed on cleanup errors

cursoragent
and
rchalamala
committed
test(sglang): require symmetric evidence in full TP matrix

cursoragent
and
rchalamala
committed
fix(sglang): stabilize concurrent TP oracle

cursoragent
and
rchalamala
committed
fix(sglang): force native concurrent batch

cursoragent
and
rchalamala
committed
fix(sglang): guarantee native scheduler batching

cursoragent
and
rchalamala
committed
test(sglang): align native batch decode lifetimes

cursoragent
and
rchalamala
committed
fix(sglang): pin decode_log_interval=1 so bs>1 replay stays observable

cursoragent
and
rchalamala
committed
docs(sglang): record verified symmetric TP evidence

cursoragent
and
rchalamala
committed
docs(sglang): address symmetric TP review findings

cursoragent
and
rchalamala
committed
docs(sglang): clarify LOAD metadata planning

cursoragent
and
rchalamala
committed

Commits on Jul 24, 2026

docs: plan final branch deltas

cursoragent
and
rchalamala
committed
fix: harden VMM IPC process and reservation ownership

cursoragent
and
rchalamala
committed
fix(ipc): harden VMM-IPC ownership claim and reservation cleanup

cursoragent
and
rchalamala
committed
test(ipc): expand fork rejection and add native lifecycle regressions

cursoragent
and
rchalamala
committed
fix(ipc): claim ownership first, assert lock-free pid, log fallback free

cursoragent
and
rchalamala
committed
test(ipc): cross-process stale exporter, VA-unreserved and same-VA proofs

cursoragent
and
rchalamala
committed
fix(ipc): pure generation-sensitive chunk key + fail-closed fallback free

cursoragent
and
rchalamala
committed
test(ipc): observable generation recreate + fully bounded lifecycle waits

cursoragent
and
rchalamala
committed
fix(ipc): correct SPDX header comment style in new C++ files

cursoragent
and
rchalamala
committed
test(ipc): generation recreate uses one region, distinct base per generation

cursoragent
and
rchalamala
committed
test(ipc): drain caching allocator before freeing each generation

cursoragent
and
rchalamala
committed
feat(vllm): archive DeepEP transport profile

cursoragent
and
rchalamala
committed
fix(vllm): validate DeepEP transport profile early

cursoragent
and
rchalamala
committed
docs(vllm): define no-IMEX DeepEP transport

cursoragent
and
rchalamala
committed
fix(vllm): align strict DeepEP matrix recipes

cursoragent
and
rchalamala
committed
test: require exact restored DeepEP output

cursoragent
and
rchalamala
committed
fix(test): harden DeepEP matrix orchestration

cursoragent
and
rchalamala
committed
test(vllm): make TP archive publication transactional

cursoragent
and
rchalamala
committed
feat(test): reject symlinks in TP archive inventory

cursoragent
and
rchalamala
committed
fix(test): make vLLM TP archive publication auto-commit safe

cursoragent
and
rchalamala
committed
test(vllm): assert LOAD live-offset emitter is exactly-once, LOAD-only

cursoragent
and
rchalamala
committed
fix(test): reject a symlinked TP archive root

cursoragent
and
rchalamala
committed
fix(test): harden vLLM TP publication and transaction validation

cursoragent
and
rchalamala
committed
fix(test): strict TP transport profile and portable manifest-gated publish

cursoragent
and
rchalamala
committed
fix(test): make TP publication authority a Modal Dict, not on-volume markers

cursoragent
and
rchalamala
committed
vLLM TP: atomic self-reconciling publication marker, bounded record, LOAD reload

cursoragent
and
rchalamala
committed
vLLM TP: strict canonical cleanup, full record validation, wrapped reconciliation

cursoragent
and
rchalamala
committed
vLLM TP: deep payload/snapshot validation, on-volume byte re-check, idempotent cleanup retry

cursoragent
and
rchalamala
committed
vLLM TP: idempotent cleanup for lost pop ack; fail closed on unindexed orphan

cursoragent
and
rchalamala
committed
test(vllm): complete TP restart acceptance

cursoragent
and
rchalamala
committed
test(vllm): repair DeepEP fixture isolation

cursoragent
and
rchalamala
committed
test(vllm): close TP review gaps

cursoragent
and
rchalamala
committed
feat(tp-helpers): optional exact descriptor-profile validation

cursoragent
and
rchalamala
committed
fix(vllm-tp): seed is diagnostic; save/save2 enforce exact descriptor set

cursoragent
and
rchalamala
committed
fix(vllm-tp): descending descriptor order, realistic seed66, exact reduced schema

cursoragent
and
rchalamala
committed
test(vmm): split hooked and driver symbol bindings

cursoragent
and
rchalamala
committed
test(vmm): keep address probes hook-interposed

cursoragent
and
rchalamala
committed
fix(test): allow uv system install in DeepEP image

cursoragent
and
rchalamala
committed
test(deepep): enforce exact routing replay oracle

cursoragent
and
rchalamala
committed
fix(test): load graphs without saved outputs

cursoragent
and
rchalamala
committed
docs: record final vLLM TP validation

cursoragent
and
rchalamala
committed
docs: correct final DeepEP evidence

cursoragent
and
rchalamala
committed
docs: clarify validation exit codes

cursoragent
and
rchalamala
committed
fix(vmm): make cleanup failures retryable

cursoragent
and
rchalamala
committed
fix(sglang): fail closed on preallocation errors

cursoragent
and
rchalamala
committed
fix(test): report SGLang Volume failures conservatively

cursoragent
and
rchalamala
committed
chore(vllm): normalize imports and cleanup docs

cursoragent
and
rchalamala
committed
fix(vmm): make allocation cleanup retryable

cursoragent
and
rchalamala
committed
fix(sglang): validate preallocation teardown state

cursoragent
and
rchalamala
committed
docs: align cleanup and retention contracts

cursoragent
and
rchalamala
committed
style(test): format VMM source contract

cursoragent
and
rchalamala
committed
fix(vmm): retain orphaned import cleanup

cursoragent
and
rchalamala
committed
docs: record final unified runtime evidence

cursoragent
and
rchalamala
committed
docs: record final local validation count

cursoragent
and
rchalamala
committed

Footer

© 2026 GitHub, Inc.

Footer navigation

Terms
Privacy
Security
Status
Community
Docs
Contact

You can’t perform that action at this time.