[ROCm][CI] Transformers: pass only one of input_ids/inputs_embeds (#48258)

stefankoncarevic committed 4 months ago

[EC Connector] CPU Offloading EC Connector (#47423)

omerpaz95 committed 4 months ago

[Misc] Improve Matryoshka pooling dimensions validation (#48057)

Taneem Ibrahim committed 4 months ago

[BugFix][ModelRunner V2] Fix stale attn metadata in speculator prefill cudagraph capture (#48261)

Nick Hill committed 4 months ago

[XPU][CI]Adjust timeout_in_minutes in Intel GPU CI (#48418)

xiangdong committed 4 months ago

remove force channels_last in Idefics3MultiModalProcessor (#48467)

Yan Ma committed 4 months ago

lower memory required for capturing cudagraphs for large cudagraph sizes (#48483)

Omer Ullman Argov committed 4 months ago

[Mypy Fix] Split mypy work (#48490)

Wentao Ye committed 4 months ago

[CI] Add SPDX license header to Rust/Protobuf sources (#48472)

Bugen Zhao committed 4 months ago

[Core] Support fp32 lm_head for generation models via head_dtype (RFC #48305 §3.6) (#48390)

Karthik Kothuri committed 4 months ago

[Bugfix][KV Cache] Don't route uniform-page-size MLA+SWA models into DeepseekV4 packing (#48256)

Nicolò Lucchesi committed 4 months ago

[Attention] Make sliding-window support an explicit backend capability (#48011)

Nicolò Lucchesi committed 4 months ago

[Distributed][Perf] Enable FlashInfer MNNVL allreduce RMS quant fusion (#48064)

Mohammad Miadh Angkad committed 4 months ago

[ROCm][CI] Cache Rust builds by source inputs (#46527)

Andreas Karatzas committed 4 months ago

[BugFix] Restore full tokens for Qwen MTP When MoE SP (#48429)

Canlin Guo committed 4 months ago

[CPU] Create Proper Numa topology for s390x (#40714)

Rehan Khan committed 4 months ago

[Bugfix][UT]Fix EagleMiniCPMForCausalLM meet TypeError (#48452)

Yejing Lai committed 4 months ago

[CPU][Spec Decode] Support DFlash speculative decoding for GDN models on CPU (#46090)

guybd committed 4 months ago

Re-disable CUDA graph memory profiling on ROCm (#48440)

Rohan Potdar committed 4 months ago

[Bugfix][ROCm] Keep TP all_gather on base-class collective (#48446)

Fangzhou Ai committed 4 months ago