refactor(runtime)!: adopt the modern Infini stack - #506

Draft

voltjia wants to merge 23 commits into

mainfrom

refactor/adopt-modern-infini-stack

Commits on Aug 13, 2026

refactor(runtime)!: adopt modern Infini stack
voltjia
committed
e7a21ba
docs: define Infini stack repository boundaries
voltjia
committed
22ccc47
refactor: own Infini stack integration build
voltjia
committed
c1d7d1d
docs: document the InfiniLM stack build
voltjia
committed
5a76190
ci: validate the modern NVIDIA stack
voltjia
committed
0700e5a
fix: keep static graph cache metadata dynamic
voltjia
committed
7611379
refactor(ops): use canonical InfiniOps APIs
voltjia
committed
bcf4054
fix(build): select required linked InfiniOps providers
voltjia
committed
235c234
fix(build): limit compiled models to Qwen3
voltjia
committed
b2ab7c1
feat(distributed): add eager point-to-point wrappers
voltjia
committed
7785bb1
feat(ops): restore prepacked linear execution
voltjia
committed
173029c
fix(ops): match canonical Gemm call schema
voltjia
committed
468828f
fix(sampling): support scalar sample outputs
voltjia
committed
d97cc45
fix(nn): pass RoPE output handles by value
voltjia
committed
878b950
fix(attention): include variable-length MHA declaration
voltjia
committed
5b12f21
fix(engine): use local static graph assertions
voltjia
committed
1cfed1b
fix(config): reject unsupported Qwen3 linear bias
voltjia
committed
c2d8671
test: cover modern runtime contracts
voltjia
committed
821d230
docs: update modern Qwen3 support boundary
voltjia
committed
febfe97
refactor(build): select InfiniOps implementations from JSON
voltjia
committed
077867b
style: format InfiniOps attention adapters
voltjia
committed
e671313
fix(graph): update paged replay metadata through tensors
voltjia
committed
ec1f6c2
fix(graph): fall back to eager paged decode under tensor parallelism
voltjia
committed
9439ea6

Navigation Menu

Uh oh!

refactor(runtime)!: adopt the modern Infini stack - #506

refactor(runtime)!: adopt the modern Infini stack#506
voltjia wants to merge 23 commits into
mainfrom
refactor/adopt-modern-infini-stack

Commits on Aug 13, 2026

refactor(runtime)!: adopt modern Infini stack

docs: define Infini stack repository boundaries

refactor: own Infini stack integration build

docs: document the InfiniLM stack build

ci: validate the modern NVIDIA stack

fix: keep static graph cache metadata dynamic

refactor(ops): use canonical InfiniOps APIs

fix(build): select required linked InfiniOps providers

fix(build): limit compiled models to Qwen3

feat(distributed): add eager point-to-point wrappers

feat(ops): restore prepacked linear execution

fix(ops): match canonical Gemm call schema

fix(sampling): support scalar sample outputs

fix(nn): pass RoPE output handles by value

fix(attention): include variable-length MHA declaration

fix(engine): use local static graph assertions

fix(config): reject unsupported Qwen3 linear bias

test: cover modern runtime contracts

docs: update modern Qwen3 support boundary

refactor(build): select InfiniOps implementations from JSON

style: format InfiniOps attention adapters

fix(graph): update paged replay metadata through tensors

fix(graph): fall back to eager paged decode under tensor parallelism