Eagle3 that supports the Minicpm3 model (#24243)

阿丹(adan) committed about 1 year ago

[Misc] fix unique_filepath (#25732)

Jiangyun Zhu committed about 1 year ago

[ray][metrics] Replace ':' with '_' for OpenTelemetry compatibility in Ray (#25439)

Seiji Eicher committed about 1 year ago

[BugFix] Fix using `dbo_decode_token_threshold` always (and ignoring `dbo_prefill_token_threshold`) (#25622)

Lucas Wilkinson committed about 1 year ago

[Quantization] Add field to skip unquantized modules for GPTQ config (#25455)

Isotr0py committed about 1 year ago

[CI/Build] Fix some V1 tests not being run (#25569)

Cyrus Leung committed about 1 year ago

[CI/Build] Split up Distributed Tests (#25572)

Cyrus Leung committed about 1 year ago

[Bugfix] Properly abort pooling request. (#25734)

wang.yuqi committed about 1 year ago

[CI] Fix test_shared_storage_connector_hashes (#25748)

Chauncey committed about 1 year ago

[Model] Mamba2 varlen refactor (#21467)

Chih-Chieh Yang committed about 1 year ago

[Doc] Update Batch-level DP docs (#25757)

Cyrus Leung committed about 1 year ago

Support LongCat-Flash-Chat tool call (#24083)

Xu Wenqing committed about 1 year ago

[Bugfix] Fix Shared Expert/Zero expert code in FusedMoE.process_chunk (#25698)

Sage Moore committed about 1 year ago

[misc] refactor speculative config (#25657)

yyzxw committed about 1 year ago

Remove cuda hard-code in compute_causal_conv1d_metadata (#25555)

Icey committed about 1 year ago

[Qwen3-Next][GDN] fixes cuda graph capturing bug in GDN metadata and a stride bug in causal_conv_1d. (#25743)

Tao He committed about 1 year ago

fix: print outputt offline_inference/base/chat.py example (#25744)

Iceber Gu committed about 1 year ago

[Harware][AMD][Model] Triton MoE tuning configs for GLM-4.5 for MI300X (#25703)

xaguilar-amd committed about 1 year ago

perf: Avoid copying inputs_embeds tensors to GPU unless prompt_embeds is enabled (#25739)

Andrew Sansom committed about 1 year ago

fix: revert cast to cpu in `MsgpackEncoder._encode_tensor` to avoid hidden performance regressions (#25738)

Andrew Sansom committed about 1 year ago