Issues

• Invitation to review Fast-dDrive in the VLA4AD survey collection

  NVlabs/Fast-dLLM#84 · Filter by author ldkong1205 opened on Oct 14, 2026

• Questions about use_block_cache mechanism in block diffusion generation

  NVlabs/Fast-dLLM#83 · Filter by author Wxp01 opened on Sep 18, 2026

• No speedup on short generation (~20–30 tokens) with Fast_dLLM_v2_1.5B (threshold=0.9)

  NVlabs/Fast-dLLM#81 · Filter by author ShallowMDream opened on Aug 14, 2026

• Why does MBPP not require post-processing while HumanEval does?

  NVlabs/Fast-dLLM#73 · Filter by author Sugar929 opened on Mar 18, 2026

• Training instability and gradient explosion when finetuning Fast_dLLM_v2_7B on Alpaca dataset

  NVlabs/Fast-dLLM#69 · Filter by author longxiangwang opened on Feb 4, 2026
  1

• Packing SFT Data

  NVlabs/Fast-dLLM#68 · Filter by author mostafaelhoushi opened on Jan 31, 2026

• 为什么dual cache模式下，humaneval的准确率为0

  NVlabs/Fast-dLLM#66 · Filter by author zqc214 opened on Jan 17, 2026
  2

• Batch Inference using Fast-dLLM

  NVlabs/Fast-dLLM#65 · Filter by author Gaurav7888 opened on Jan 16, 2026
  1

• batch size > 1 and temperature != 0.0 not supported by the generate function file

  NVlabs/Fast-dLLM#64 · Filter by author Laurence-Wu opened on Jan 16, 2026

• Discrepancy in GSM8K evaluation results compared to Table 1

  NVlabs/Fast-dLLM#63 · Filter by author JueVioletGrace opened on Jan 11, 2026

• [v2] Reproducibility: release paper training script, preprocessing, and exact hyperparameters

  NVlabs/Fast-dLLM#62 · Filter by author yeonjoon-jung01 opened on Jan 5, 2026
  1

• Why does Dream’s cache “global update” always update from current_block_start instead of using a sampling strategy?

  NVlabs/Fast-dLLM#61 · Filter by author lizhuo-luo opened on Dec 30, 2025

• (v2) Open source code for transferring AR model to dLLM?

  NVlabs/Fast-dLLM#60 · Filter by author Xalp opened on Dec 30, 2025
  1

• 请问为什么在eval_gsm8k.sh中，dual cache+paralle的steps参数是传入length，而不是steps

  NVlabs/Fast-dLLM#59 · Filter by author zqc214 opened on Dec 29, 2025
  1

• do we need chat template when evaluating llada-1.5

  NVlabs/Fast-dLLM#58 · Filter by author mynameisissac opened on Dec 28, 2025

• Question Regarding the Llada Parallel decoding in GSM8K

  NVlabs/Fast-dLLM#54 · Filter by author littlestone111 opened on Dec 18, 2025
  2

• About warm-up stage and cache reuse stage in dual cache+parallel

  NVlabs/Fast-dLLM#52 · Filter by author quaternior opened on Dec 11, 2025

• vllm support

  NVlabs/Fast-dLLM#50 · Filter by author hsliuustc0106 opened on Dec 8, 2025
  1

• About tokens per second

  NVlabs/Fast-dLLM#49 · Filter by author xiaoshideta opened on Dec 2, 2025
  1

• It seems that the cfg_scale is removed?

  NVlabs/Fast-dLLM#44 · Filter by author yjyddq opened on Nov 17, 2025
  1

• Completion of questions under multiple batch sizes?

  NVlabs/Fast-dLLM#40 · Filter by author zs1314 opened on Nov 5, 2025

• Performance differences caused by batch_size

  NVlabs/Fast-dLLM#39 · Filter by author zs1314 opened on Nov 5, 2025
  1

• fast-dllm-v2中，dual cache的实际作用？

  NVlabs/Fast-dLLM#36 · Filter by author mZhenz opened on Oct 30, 2025
  6

• fast-dllm-v2 mbpp and humaneval evaluation

  NVlabs/Fast-dLLM#34 · Filter by author Connor323 opened on Oct 29, 2025
  1

• [Feature Request] Support for LLaDAMoE model

  NVlabs/Fast-dLLM#28 · Filter by author Stanleytowne opened on Oct 11, 2025