Sunbelt Computer Software

PL/B Language Development and Support

History for max/python - modular/modular · GitHub

Commits

History for

Commits on Jun 26, 2026

[MAX] Migrate text LLM architectures to batch processor

cadddr
authored and
modularbot
committed
[MAX] Migrate DeepSeek architectures to batch processor

cadddr
authored and
modularbot
committed
[MAX] Add Gemma-4 MoE NVFP4 fp8kv smoke recipe

hsinyuting
authored and
modularbot
committed
[stdlib] support stdlib_plugin_name attr on kgen.target

szbergeron
authored and
modularbot
committed
[MAX][Models] Cache Gemma4 video vision-encoder outputs

bhansconnect
authored and
modularbot
committed
[Kernels] Add grouped allgather support

lshzh-ww
authored and
modularbot
committed
[Serve] Improve multimodal preprocessor time-to-first-token

rowland-208
authored and
modularbot
committed
[MAX] Delete some dead code, NFC

atomicapple0
authored and
modularbot
committed
[driver] Restructure hal interfaces

szbergeron
authored and
modularbot
committed
[Kernels] Add grouped reduce-scatter support

lshzh-ww
authored and
modularbot
committed
[MAX] Add reasoning_split request field for MiniMax M3

rowland-208
authored and
modularbot
committed
[MAX] Move serve optimization flags to SupportedArchitecture (#89882)

katelyncaldwell
authored and
modularbot
committed
[Docs] Make experimental docstring examples executable and verified

rachfop
authored and
modularbot
committed
[MAX] Fix disagg decode bitmask callback crash on KV-transferred rows

AustinDoolittle
authored and
modularbot
committed
[stdlib] Parameterize HAL `Buffer` by `device_spec`

JoeLoser
authored and
modularbot
committed

Commits on Jun 25, 2026

[MAX] Default eager interpreter to lazy compilation; add

pprovins
authored and
modularbot
committed
[ConstrainedDecoding] Accept pinned-host bitmask in xgrammar binding

shaurya0
authored and
modularbot
committed
[ConstrainedDecoding] Enforce Kimi tool-call args via xgrammar

shaurya0
authored and
modularbot
committed
[MAX] experimental: trim the API docs to the public surface

TilliFe
authored and
modularbot
committed
[ConstrainedDecoding] Add pluggable grammar backend

shaurya0
authored and
modularbot
committed
[KVCache] Support mixed DP and TP KV cache

lshzh-ww
authored and
modularbot
committed
[bazel] Delete the `//max/python/max:max` target

Ahajha
authored and
modularbot
committed
[lint] Fix various typos

Ahajha
authored and
modularbot
committed
[MAX] Split KVCacheParams into MHA/MLA/MSA subclasses

atomicapple0
authored and
modularbot
committed
[MAX] Trim down dependencies of `max benchmark`

Ahajha
authored and
modularbot
committed
[Config] Deep-merge config-file values with CLI/env overrides

tzhenghao
authored and
modularbot
committed
[Models][Kernels] Full 5-way fused QKV+index projection for MiniMax-M3

hamid-74
authored and
modularbot
committed
[MAX][Models] Add video content hashes to KV-cache keys

JasxnNg
authored and
modularbot
committed
[MAX] Make `max.entrypoints` private

Ahajha
authored and
modularbot
committed
[MAX] Normalize Hugging Face model IDs to canonical casing

atomicapple0
authored and
modularbot
committed
[KVConnector] Shard disk-tier cache into per-hash subdirectories

atomicapple0
authored and
modularbot
committed
[mojo] Remove parameter-list "where" IR support

zyx-billy
authored and
modularbot
committed
[KVConnector] Evict disk-tier blocks off the scheduler critical path

atomicapple0
authored and
modularbot
committed
[MAX][Models] Fix Gemma4 chunked-prefill video scatter indices

bhansconnect
authored and
modularbot
committed
[Models][CI] Set K2.6 Eagle3 draft sliding_window in smoke-test recipes

csk7
authored and
modularbot
committed