Oneflow-Inc/libai

Issues

• GPT2预训练模型，相同配置下libai的显存占用率会显著高于megatron-lm

  #349 opened 2 years ago by Sakura-gh
  11

• LLaMA-7B SFT died with <Signals.SIGABRT: 6>

  #539 opened 8 months ago by PussyCat0700
  0

• 单机多卡跑gpt2_pretrain.py遇到如下问题

  #534 opened 10 months ago by treestreamymw
  0

• 运行GLM示例报错 module 'oneflow._C' has no attribute 'fused_multi_head_attention_inference_v2'

  #512 opened 2 years ago by bmfire1
  1

• 建议requirements 中涉及requests指定一下具体版本

  #522 opened a year ago by digger-yu
  0

• Project下的MAE多卡训练报错

  #508 opened 2 years ago by stonewjf
  0

• 运行教程的bash tools/train.sh tools/train_net.py configs/vit_imagenet.py 8 命令报错

  #498 opened 2 years ago by stonewjf
  0

• GLM 10B CN推理加速耗时

  #479 opened 2 years ago by vicwer
  1

• 测试并行框架，张量并行结果与官网所给数据不一致

  #477 opened 2 years ago by lisuq
  0

• GPT2预训练，libai的throughput和以前的数据不匹配

  #475 opened 2 years ago by lisuq
  1

• [多机多卡][MT5]failed to connect to all addresses

  #470 opened 2 years ago by MikeDean2367
  1

• MT5和T5的区别

  #468 opened 2 years ago by MikeDean2367
  4

• GLM libai推理报错

  #464 opened 2 years ago by tanklandry
  2

• TypeError: __init__() got an unexpected keyword argument 'flags'

  #463 opened 2 years ago by michelleqyhqyh
  5

• 微信群满了

  #433 opened 2 years ago by MissiontoMars
  3

• 纯tensor并行训练，4卡和8卡使用的集合通信算子不同

  #455 opened 2 years ago by Panlichen
  2

• 设置不同的data_parallel_size导致了不同的global_batch_size

  #345 opened 2 years ago by Yipeng1994
  1

• CI test 失效

  #435 opened 2 years ago by leaves-zwx
  0

• [MT5] exec_graph.cpp physical shape check failed.

  #405 opened 2 years ago by strint
  2

• [Bug]libai test error：File exists: './data_test/bert_data'

  #428 opened 2 years ago by strint
  3

• 关于benchmark实验结果的疑问

  #421 opened 2 years ago by frankxyy
  2

• [Bug][MT5] Throughput is unexpected

  #406 opened 2 years ago by strint
  47

• 多机训练失败后，非master node的进程没有完全kill掉

  #416 opened 2 years ago by frankxyy
  3

• 多机训练报错

  #415 opened 2 years ago by frankxyy
  13

• [Bug][MT5] graph compile time

  #407 opened 2 years ago by strint
  27

• MT5 8卡纯模型并行，graph模式运行报错

  #409 opened 2 years ago by Ldpe2G
  12

• 可否支持读取pytorch model进行训练

  #414 opened 2 years ago by frankxyy
  3

• python requirements缺失？

  #413 opened 2 years ago by frankxyy
  0

• [Bug][MT5]在特定环境下出现 build model 时间很久的问题

  #404 opened 2 years ago by strint
  25

• Swin Transformer V2 基于LiBai复现 (关于CCF的pr整合)

  #359 opened 2 years ago by shaoshitong
  2

• PyTorchJob 的方式在 Kubernetes 多机多卡启动 libai 执行

  #368 opened 2 years ago by strint
  1

• 我在复现DETR代码时，在并行计算时出现问题，请问该怎么解决。

  #376 opened 2 years ago by oooo111
  0

• 基于libai复现SegFormer[projects]

  #342 opened 2 years ago by zhanggj821
  7

• [WIP] Swin 和 Swin v2 训练精度对齐官方实现

  #357 opened 2 years ago by Ldpe2G
  0

• DETR结果对齐实验记录

  #288 opened 3 years ago by HiHippie
  25

• 运行tools/train.sh脚本报错：Check failed: num_device > 0 (0 vs. 0) No IB device found

  #340 opened 2 years ago by Sakura-gh
  3

• [TODO] DeiT III 的训练策略同步

  #266 opened 2 years ago by rentainhe
  1

• Module重构讨论

  #335 opened 2 years ago by xiezipeng-ML
  7

• swin 数据并行，8卡线性加速比，比pytorch低

  #312 opened 3 years ago by Ldpe2G
  6

• Swin 数据并行从单进程到8进程，显存暴涨

  #311 opened 3 years ago by Ldpe2G
  8

• LiBai Benchmark README.md

  #317 opened 3 years ago by chengtbf
  0

• 想问问现在有转换hugging face模型的代码了吗？

  #310 opened 3 years ago by drxmy
  5

• 多卡模型加载 key 异常

  #293 opened 3 years ago by strint
  0

• Swin 模型加载 checkpoint 训练0号卡显存显著增多

  #292 opened 3 years ago by Ldpe2G
  28

• [Research] Support ffcv backend

  #280 opened 3 years ago by rentainhe
  1

• 期望libai mae支持graph格式数据并行，流水线并行和模型并行

  #259 opened 3 years ago by KellyZhang2020
  5

• 推理和生成相关调研和设计

  #265 opened 3 years ago by L1aoXingyu
  7

• 流水并行下 get_batch 中 data to_global 的修复方式讨论

  #257 opened 3 years ago by Ldpe2G
  1

• [BUG] mae finetune 加载官方mae vit模型失败

  #258 opened 3 years ago by KellyZhang2020
  1

• Documentation Guide

  #246 opened 3 years ago by rentainhe
  0