LoreonLabsPlatform
DocsHome
  • Overview

Intelligence

  • Markets
  • Builders
  • Research
  • Ecosystems
  • Launchpads
  • Search
Ecosystems

Other

SpecForge

Train speculative decoding models effortlessly and port them smoothly to SGLang serving.

OtherEmerging
GitHubWebsite
Stars
—
Forks
—
Contributors
8
Last push
8mo ago

Recent commits

Latest commits.

  • Feature/add sgl online (#239)
    49452e2zyksir9mo ago
  • [feat]Regenerate train data from target model. (#223)
    a2cb6d8Ximingwang-099mo ago
  • fix: handle None test_ds in skipped messages count (#232)
    07157bdLancer9mo ago
  • docs: run build_eagle3_dataset.py with torchrun (#234)
    ec6f0d9Bao Jiangnan9mo ago
  • Optimize Evaluation Phase By adding torch.no_grad() (#237)
    3f0ae30Ximingwang-099mo ago
Fix FP32→BF16 redundant allocation during model init (#228)
ca2fc85ggg-s9mo ago
  • Replace unsupported --batch-size with --draft-micro-batch-size and --… (#238)
    d562e50xinjiahui9mo ago
  • Offline Train: Avoid Duplicate Model Download (#235)
    20406f4Bao Jiangnan9mo ago

    • Top contributors

    Builders behind this project.

    FrankLeeeee
    14 commits
    fzyzcjy
    11 commits
    sleepcoo
    7 commits
    Ximingwang-09
    6 commits
    yubofredwang
    6 commits
    ZhengHSI
    6 commits
    shimizust
    6 commits
    zyksir
    5 commits