GeezerVison

Pull requests

• Add a multi-turn GRPO example with tool calling, and port OPSD to the per-sample rollout API

  #1023·anyadontfly opened Nov 9, 2026·

  ·

• Add AutoTP training to rollout weight sync example

  #1021·jinyouzhi opened Oct 22, 2026Contributor·

  ·
  2

• Add AutoEP HF checkpoint to rollout precision check

  #1020·jinyouzhi opened Oct 21, 2026Contributor·

  ·

• Add vocabulary-parallel KL support to OPSD

  #1019·0z5a opened Oct 14, 2026·

  ·
  3

• benchmarks: move adaptive prefill calibration to Examples

  #1018·SuperGoodGame opened Oct 10, 2026·

  ·
  19

• Add ReFlow ZeRO-3 CPU-offload fine-tuning examples

  #1017·st-bang97 opened Oct 2, 2026·

  ·

• Perf/opsd continuous batching benchmark clean

  #1011·nathon-lee opened Aug 30, 2026Contributor·

  ·

• perf(opsd): benchmark shared prefill with CUDA Graph decode

  #1010·nathon-lee opened Aug 29, 2026Contributor·

  ·

• Fix PPO's compute_reward() (#996)

  #997·CZH-THU opened Dec 22, 2025·

  ·

• support loading local models

  #993·KairongChen-l opened Dec 9, 2025·

  ·

• Revert "fix init weights issue for critic/reward model (#983)"

  #985·hwchen2017 opened Jul 29, 2025Contributor·

  ·

• add fused linear-loss function in Domino

  #965·duanhx1037 opened Apr 8, 2025·

  ·
  1

• [Draft] Add support for seq split in Domino

  #961·duanhx1037 opened Mar 4, 2025·

  ·
  3

• Update Domino for Llama3

  #959·shenzheyu opened Feb 26, 2025Contributor·

  ·
  1

• add checkpoint

  #945·zhangsmallshark opened Dec 16, 2024Contributor·

  ·
  20

• Bump the pip group across 9 directories with 15 updates #3

  #938·akaday opened Nov 25, 2024·

  ·
  1

• Add LoRA optimization to the SD training example

  #873·PareesaMS opened Mar 8, 2024Contributor·

  ·

• [DeepSpeed-Chat] Fix OOM issue in dataloader

  #841·youkaichao opened Jan 1, 2024·

  ·
  4

• fix: the source part should not participate in loss calculation in SFT stage

  #762·xffxff opened Oct 10, 2023Contributor·

  ·
  1

  1

• Inference test enhance

  #713·sakogan opened Aug 31, 2023·

  ·
  1

• Add FALCON inference-test

  #557·RezaYazdaniAminabadi opened May 30, 2023Contributor·

  ·

• [DeepSpeed-Chat] feat: Support local path for actor, reward models.

  #485·sepilqi opened May 6, 2023·

  ·
  2

• Fix chatbot

  #462·yaozhewei opened Apr 28, 2023Contributor·

  ·

• fix step_time

  #454·thuzhf opened Apr 28, 2023·

  ·

• Fix numerical instability of critic loss

  #424·s-isaev opened Apr 25, 2023Contributor·

  ·