Commits
User selector
Commit history
Commits on Jul 28, 2026
[KVCache][NIXL] Split generic TransferEngine + metadata bases and extract pure topology resolver
Show description for 68887c0authored andcommitted[KV] Warn at startup about leftover `max_kv_tiered_*` offload
Show description for 5ee42b8authored andcommitted[max] slightly optimize sha256
Show description for 9d4c5b7authored andcommitted[MAX] Bound vision encoding to the chunked-prefill active window
Show description for c50d9b5authored andcommitted[KVConnector] Default the tiered and rust_tiered connectors' disk worker
Show description for 9d08903authored andcommitted[max] shrink overflowed vision cache asap
Show description for 7031569authored andcommitted[MAX] Route eager norm ops (layer_norm/rms_norm/group_norm) through the
Show description for 50a0611authored andcommitted[cascade] Optimizations
Show description for fffd89eauthored andcommitted[MAX] Sample recovered tokens from the residual in stochastic acceptance (#93233)
Show description for 6741c40authored andcommitted[MAX][Profiling] Refactor the MAX profiler into a runtime-loaded plugin
Show description for 75e873cauthored andcommitted[MAX] Batch multi-device replay preface copies into one submission
Show description for 6a039caauthored andcommitted[Docs] Standardize graph/functional/tensor op docstrings
Show description for b5a61b2authored andcommitted[Kernels][MAX] Fused all-gather + RMSNorm op, gated MiniMax-M3 wiring
Show description for 9e34fd4authored andcommitted[dKV] surfaces dKV connector connection stats through KVCacheMetrics and
Show description for 9dafe99authored andcommitted[MAX] Fix DI vision-context handling gaps
Show description for 71371c1authored andcommitted
Commits on Jul 27, 2026
[MAX] Move VLM vision-merge input finalization into the pipeline seam
Show description for c578b57authored andcommitted
Commits on Jul 26, 2026
[stdlib] Rename InlineArray to Array in remaining areas (#93377)
Show description for e608e45authored andcommitted[Kernels] Fix mis-aligned input load in rms_norm reduction kernels
Show description for 6f3d2a5authored andcommitted[Pipelines] Fix CLI kv-cache flags clobbering config-file values
Show description for 5fbb062authored andcommitted[MAX] Retry HF weight and dataset downloads past a racy `.incomplete`
Show description for e09be27authored andcommitted
Commits on Jul 25, 2026
[MAX] Register EOS tokens with XGrammar
Show description for 8839a6fauthored andcommitted[MAX][NFC] Fold tokenizer EOS into a single public eos_token_ids set
Show description for d15349fauthored andcommitted[Pipelines] Re-resolve weight-path identity in
Show description for d1ac613authored andcommitted[MAX] Fix Kimi-K2.5 device graph capture after vision cache migration
Show description for a24d59cauthored andcommitted[MAX] Constrain response format with auto tool choice
Show description for 460322cauthored andcommitted[mojo-lang] Support custom messages on `where` clauses
Show description for f78847bauthored andcommitted[cascade] We need a `LocalRustRuntime` in Rust to do proper
Show description for fdfd487authored andcommitted[stdlib] Rename InlineArray to Array (#93254)
Show description for 08a8c9cauthored andcommitted[KV] Migrate smoke tests from tiered/local to rust_tiered KV connector
Show description for 1d0f585authored andcommitted[MAX] Emit compact JSON for structured output
Show description for 744cf23authored andcommitted[Pipelines] Wire the fused LoRA-B QKV expand through the LoRA pipeline
Show description for 868bc6bauthored andcommitted[stdlib] Have `InlineArray.unsafe_ptr` return a safe `Pointer`
Show description for e636d29authored andcommitted[KV] Adds AMD (HIP) support to the rust_tiered KV connector.
Show description for 14903bcauthored andcommitted[MAX][NFC] Make Gemma 4 structural tag builder public
Show description for 9384cc1authored andcommitted[MAX] Let speculative decoding run to max_seq_len instead of
Show description for 2e7554eauthored andcommitted