Segmentation Performance Evaluation Result is significantly different than paper gIoU:60.0 cIoU:67.8 vs gIoU:0.0759 cIoU:0.0648
ssablak opened this issue · comments
Hi @X-Lai, @tianzhuotao, @yukang2017, @yanwei-li , @xbkaishui
I am evaluating your method with your provided model (xinlai/LISA-13B-llama2-v1) as a part of my research studies. I see the significant difference between your paper and the evaluation result xinlai/LISA-13B-llama2-v1 by utilizing your --eval_only flag.
your paper reports as below:
LISA-Llama2-13B (ft) gIoU: 60.0 cIoU: 67.8
When I run your source code with --eval_only
LISA-Llama2-13B (ft) gIoU: 0.0759 cIoU: 0.0648
Could you please help me with what I am missing? or what is the best way to evaluate your released model?
Thanks,
-Steve
(test) ssablak@ssablak-5820u:/media/$ python train_ds.py --version="/media/models/LISA-13B-llama2-v1" --dataset_dir="/media/dataset" --vision_pretrained="/media/models/SAM/sam_vit_h_4b8939.pth" --exp_name="LISA-13B-llama2-v1" --eval_only
[2024-01-09 18:49:34,938] [INFO] [real_accelerator.py:161:get_accelerator] Setting ds_accelerator to cuda (auto detect)
You are using the legacy behaviour of the <class 'transformers.models.llama.tokenization_llama.LlamaTokenizer'>. This means that tokens that come after special tokens will not be properly handled. We recommend you to read the related pull request available at huggingface/transformers#24565
Loading checkpoint shards: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:44<00:00, 14.91s/it]
trainable params: 6,553,600 || all params: 13,999,497,520 || trainable%: 0.04681310876077786
n: base_model.model.model.embed_tokens.weight p.shape: torch.Size([32003, 5120])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.self_attn.q_proj.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.self_attn.q_proj.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.self_attn.k_proj.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.self_attn.k_proj.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.self_attn.v_proj.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.self_attn.v_proj.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.self_attn.out_proj.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.self_attn.out_proj.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.norm1.weight p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.norm1.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.cross_attn_token_to_image.q_proj.weight p.shape: torch.Size([128, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.cross_attn_token_to_image.q_proj.bias p.shape: torch.Size([128])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.cross_attn_token_to_image.k_proj.weight p.shape: torch.Size([128, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.cross_attn_token_to_image.k_proj.bias p.shape: torch.Size([128])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.cross_attn_token_to_image.v_proj.weight p.shape: torch.Size([128, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.cross_attn_token_to_image.v_proj.bias p.shape: torch.Size([128])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.cross_attn_token_to_image.out_proj.weight p.shape: torch.Size([256, 128])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.cross_attn_token_to_image.out_proj.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.norm2.weight p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.norm2.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.mlp.lin1.weight p.shape: torch.Size([2048, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.mlp.lin1.bias p.shape: torch.Size([2048])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.mlp.lin2.weight p.shape: torch.Size([256, 2048])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.mlp.lin2.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.norm3.weight p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.norm3.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.norm4.weight p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.norm4.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.cross_attn_image_to_token.q_proj.weight p.shape: torch.Size([128, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.cross_attn_image_to_token.q_proj.bias p.shape: torch.Size([128])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.cross_attn_image_to_token.k_proj.weight p.shape: torch.Size([128, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.cross_attn_image_to_token.k_proj.bias p.shape: torch.Size([128])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.cross_attn_image_to_token.v_proj.weight p.shape: torch.Size([128, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.cross_attn_image_to_token.v_proj.bias p.shape: torch.Size([128])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.cross_attn_image_to_token.out_proj.weight p.shape: torch.Size([256, 128])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.0.cross_attn_image_to_token.out_proj.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.self_attn.q_proj.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.self_attn.q_proj.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.self_attn.k_proj.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.self_attn.k_proj.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.self_attn.v_proj.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.self_attn.v_proj.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.self_attn.out_proj.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.self_attn.out_proj.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.norm1.weight p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.norm1.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.cross_attn_token_to_image.q_proj.weight p.shape: torch.Size([128, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.cross_attn_token_to_image.q_proj.bias p.shape: torch.Size([128])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.cross_attn_token_to_image.k_proj.weight p.shape: torch.Size([128, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.cross_attn_token_to_image.k_proj.bias p.shape: torch.Size([128])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.cross_attn_token_to_image.v_proj.weight p.shape: torch.Size([128, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.cross_attn_token_to_image.v_proj.bias p.shape: torch.Size([128])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.cross_attn_token_to_image.out_proj.weight p.shape: torch.Size([256, 128])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.cross_attn_token_to_image.out_proj.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.norm2.weight p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.norm2.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.mlp.lin1.weight p.shape: torch.Size([2048, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.mlp.lin1.bias p.shape: torch.Size([2048])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.mlp.lin2.weight p.shape: torch.Size([256, 2048])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.mlp.lin2.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.norm3.weight p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.norm3.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.norm4.weight p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.norm4.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.cross_attn_image_to_token.q_proj.weight p.shape: torch.Size([128, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.cross_attn_image_to_token.q_proj.bias p.shape: torch.Size([128])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.cross_attn_image_to_token.k_proj.weight p.shape: torch.Size([128, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.cross_attn_image_to_token.k_proj.bias p.shape: torch.Size([128])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.cross_attn_image_to_token.v_proj.weight p.shape: torch.Size([128, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.cross_attn_image_to_token.v_proj.bias p.shape: torch.Size([128])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.cross_attn_image_to_token.out_proj.weight p.shape: torch.Size([256, 128])
n: base_model.model.model.visual_model.mask_decoder.transformer.layers.1.cross_attn_image_to_token.out_proj.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.final_attn_token_to_image.q_proj.weight p.shape: torch.Size([128, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.final_attn_token_to_image.q_proj.bias p.shape: torch.Size([128])
n: base_model.model.model.visual_model.mask_decoder.transformer.final_attn_token_to_image.k_proj.weight p.shape: torch.Size([128, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.final_attn_token_to_image.k_proj.bias p.shape: torch.Size([128])
n: base_model.model.model.visual_model.mask_decoder.transformer.final_attn_token_to_image.v_proj.weight p.shape: torch.Size([128, 256])
n: base_model.model.model.visual_model.mask_decoder.transformer.final_attn_token_to_image.v_proj.bias p.shape: torch.Size([128])
n: base_model.model.model.visual_model.mask_decoder.transformer.final_attn_token_to_image.out_proj.weight p.shape: torch.Size([256, 128])
n: base_model.model.model.visual_model.mask_decoder.transformer.final_attn_token_to_image.out_proj.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.norm_final_attn.weight p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.transformer.norm_final_attn.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.iou_token.weight p.shape: torch.Size([1, 256])
n: base_model.model.model.visual_model.mask_decoder.mask_tokens.weight p.shape: torch.Size([4, 256])
n: base_model.model.model.visual_model.mask_decoder.output_upscaling.0.weight p.shape: torch.Size([256, 64, 2, 2])
n: base_model.model.model.visual_model.mask_decoder.output_upscaling.0.bias p.shape: torch.Size([64])
n: base_model.model.model.visual_model.mask_decoder.output_upscaling.1.weight p.shape: torch.Size([64])
n: base_model.model.model.visual_model.mask_decoder.output_upscaling.1.bias p.shape: torch.Size([64])
n: base_model.model.model.visual_model.mask_decoder.output_upscaling.3.weight p.shape: torch.Size([64, 32, 2, 2])
n: base_model.model.model.visual_model.mask_decoder.output_upscaling.3.bias p.shape: torch.Size([32])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.0.layers.0.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.0.layers.0.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.0.layers.1.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.0.layers.1.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.0.layers.2.weight p.shape: torch.Size([32, 256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.0.layers.2.bias p.shape: torch.Size([32])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.1.layers.0.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.1.layers.0.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.1.layers.1.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.1.layers.1.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.1.layers.2.weight p.shape: torch.Size([32, 256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.1.layers.2.bias p.shape: torch.Size([32])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.2.layers.0.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.2.layers.0.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.2.layers.1.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.2.layers.1.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.2.layers.2.weight p.shape: torch.Size([32, 256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.2.layers.2.bias p.shape: torch.Size([32])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.3.layers.0.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.3.layers.0.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.3.layers.1.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.3.layers.1.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.3.layers.2.weight p.shape: torch.Size([32, 256])
n: base_model.model.model.visual_model.mask_decoder.output_hypernetworks_mlps.3.layers.2.bias p.shape: torch.Size([32])
n: base_model.model.model.visual_model.mask_decoder.iou_prediction_head.layers.0.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.iou_prediction_head.layers.0.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.iou_prediction_head.layers.1.weight p.shape: torch.Size([256, 256])
n: base_model.model.model.visual_model.mask_decoder.iou_prediction_head.layers.1.bias p.shape: torch.Size([256])
n: base_model.model.model.visual_model.mask_decoder.iou_prediction_head.layers.2.weight p.shape: torch.Size([4, 256])
n: base_model.model.model.visual_model.mask_decoder.iou_prediction_head.layers.2.bias p.shape: torch.Size([4])
n: base_model.model.model.text_hidden_fcs.0.0.weight p.shape: torch.Size([5120, 5120])
n: base_model.model.model.text_hidden_fcs.0.0.bias p.shape: torch.Size([5120])
n: base_model.model.model.text_hidden_fcs.0.2.weight p.shape: torch.Size([256, 5120])
n: base_model.model.model.text_hidden_fcs.0.2.bias p.shape: torch.Size([256])
n: base_model.model.lm_head.weight p.shape: torch.Size([32003, 5120])
ade20k: 20210
cocostuff: 118287
loading annotations into memory...
Done (t=0.70s)
creating index...
index created!
pascal_part: 4366
loading annotations into memory...
Done (t=9.70s)
creating index...
index created!
paco_lvis: 45790
mapillary: 18000
loading dataset refclef into memory...
ref_file: /media/ssablak/mogata/dataset/refer_seg/refclef/refs(unc).p
creating index...
index created.
DONE (t=3.71s)
dataset refclef (refs unc) (train split) has 17978 images and 99523 annotations.
loading dataset refcoco into memory...
ref_file: /media/ssablak/mogata/dataset/refer_seg/refcoco/refs(unc).p
creating index...
index created.
DONE (t=6.99s)
dataset refcoco (refs unc) (train split) has 16994 images and 196771 annotations.
loading dataset refcoco+ into memory...
ref_file: /media/ssablak/mogata/dataset/refer_seg/refcoco+/refs(unc).p
creating index...
index created.
DONE (t=5.24s)
dataset refcoco+ (refs unc) (train split) has 16992 images and 196737 annotations.
loading dataset refcocog into memory...
ref_file: /media/ssablak/mogata/dataset/refer_seg/refcocog/refs(umd).p
creating index...
index created.
DONE (t=7.61s)
dataset refcocog (refs umd) (train split) has 21899 images and 208960 annotations.
vqa_data: 157712
number of reason_seg samples: 239
len(self.img_to_explanation): 239
Training with 10000 examples and validating with 200 examples.
[2024-01-09 18:51:34,030] [INFO] [logging.py:96:log_dist] [Rank -1] DeepSpeed info: version=0.12.6, git-hash=unknown, git-branch=unknown
[2024-01-09 18:51:34,031] [INFO] [comm.py:637:init_distributed] cdb=None
[2024-01-09 18:51:34,031] [INFO] [comm.py:652:init_distributed] Not using the DeepSpeed or dist launchers, attempting to detect MPI environment...
[2024-01-09 18:51:35,835] [INFO] [comm.py:702:mpi_discovery] Discovered MPI settings of world_rank=0, local_rank=0, world_size=1, master_addr=192.168.5.132, master_port=29500
[2024-01-09 18:51:35,836] [INFO] [comm.py:668:init_distributed] Initializing TorchBackend in DeepSpeed with backend nccl
[W socket.cpp:426] [c10d] The server socket cannot be initialized on [::]:29500 (errno: 97 - Address family not supported by protocol).
[W socket.cpp:601] [c10d] The client socket cannot be initialized to connect to [ssablak-5820u]:29500 (errno: 97 - Address family not supported by protocol).
[W socket.cpp:601] [c10d] The client socket cannot be initialized to connect to [ssablak-5820u]:29500 (errno: 97 - Address family not supported by protocol).
[2024-01-09 18:51:41,974] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Flops Profiler Enabled: False
Using /home/ssablak/.cache/torch_extensions/py39_cu117 as PyTorch extensions root...
Detected CUDA files, patching ldflags
Emitting ninja build file /home/ssablak/.cache/torch_extensions/py39_cu117/fused_adam/build.ninja...
Building extension module fused_adam...
Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)
ninja: no work to do.
Loading extension module fused_adam...
Time to load fused_adam op: 1.52187180519104 seconds
[2024-01-09 18:51:44,440] [INFO] [logging.py:96:log_dist] [Rank 0] Using DeepSpeed Optimizer param name adamw as basic optimizer
[2024-01-09 18:51:44,440] [INFO] [logging.py:96:log_dist] [Rank 0] Removing param_group that has no 'params' in the basic Optimizer
[2024-01-09 18:51:44,753] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Basic Optimizer = FusedAdam
[2024-01-09 18:51:44,753] [INFO] [utils.py:56:is_zero_supported_optimizer] Checking ZeRO support for optimizer=FusedAdam type=<class 'deepspeed.ops.adam.fused_adam.FusedAdam'>
[2024-01-09 18:51:44,753] [INFO] [logging.py:96:log_dist] [Rank 0] Creating torch.float16 ZeRO stage 2 optimizer
[2024-01-09 18:51:44,753] [INFO] [stage_1_and_2.py:148:init] Reduce bucket size 500000000
[2024-01-09 18:51:44,753] [INFO] [stage_1_and_2.py:149:init] Allgather bucket size 500000000
[2024-01-09 18:51:44,753] [INFO] [stage_1_and_2.py:150:init] CPU Offload: False
[2024-01-09 18:51:44,753] [INFO] [stage_1_and_2.py:151:init] Round robin gradient partitioning: False
[2024-01-09 18:51:48,972] [INFO] [utils.py:791:see_memory_usage] Before initializing optimizer states
[2024-01-09 18:51:48,973] [INFO] [utils.py:792:see_memory_usage] MA 27.64 GB Max_MA 28.32 GB CA 28.46 GB Max_CA 28 GB
[2024-01-09 18:51:48,973] [INFO] [utils.py:799:see_memory_usage] CPU Virtual Memory: used = 21.61 GB, percent = 17.2%
[2024-01-09 18:51:52,356] [INFO] [utils.py:791:see_memory_usage] After initializing optimizer states
[2024-01-09 18:51:52,357] [INFO] [utils.py:792:see_memory_usage] MA 30.36 GB Max_MA 31.73 GB CA 32.55 GB Max_CA 33 GB
[2024-01-09 18:51:52,357] [INFO] [utils.py:799:see_memory_usage] CPU Virtual Memory: used = 21.61 GB, percent = 17.2%
[2024-01-09 18:51:52,357] [INFO] [stage_1_and_2.py:516:init] optimizer state initialized
[2024-01-09 18:51:55,766] [INFO] [utils.py:791:see_memory_usage] After initializing ZeRO optimizer
[2024-01-09 18:51:55,767] [INFO] [utils.py:792:see_memory_usage] MA 30.36 GB Max_MA 30.36 GB CA 32.55 GB Max_CA 33 GB
[2024-01-09 18:51:55,767] [INFO] [utils.py:799:see_memory_usage] CPU Virtual Memory: used = 21.61 GB, percent = 17.2%
[2024-01-09 18:51:55,775] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Final Optimizer = adamw
[2024-01-09 18:51:55,775] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed using configured LR scheduler = WarmupDecayLR
[2024-01-09 18:51:55,775] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed LR Scheduler = <deepspeed.runtime.lr_schedules.WarmupDecayLR object at 0x7fcf729ff460>
[2024-01-09 18:51:55,775] [INFO] [logging.py:96:log_dist] [Rank 0] step=0, skipped=0, lr=[0.0003], mom=[(0.9, 0.95)]
[2024-01-09 18:51:55,778] [INFO] [config.py:984:print] DeepSpeedEngine configuration:
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] activation_checkpointing_config {
"partition_activations": false,
"contiguous_memory_optimization": false,
"cpu_checkpointing": false,
"number_checkpoints": null,
"synchronize_checkpoint_boundary": false,
"profile": false
}
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] aio_config ................... {'block_size': 1048576, 'queue_depth': 8, 'thread_count': 1, 'single_submit': False, 'overlap_events': True}
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] amp_enabled .................. False
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] amp_params ................... False
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] autotuning_config ............ {
"enabled": false,
"start_step": null,
"end_step": null,
"metric_path": null,
"arg_mappings": null,
"metric": "throughput",
"model_info": null,
"results_dir": "autotuning_results",
"exps_dir": "autotuning_exps",
"overwrite": true,
"fast": true,
"start_profile_step": 3,
"end_profile_step": 5,
"tuner_type": "gridsearch",
"tuner_early_stopping": 5,
"tuner_num_trials": 50,
"model_info_path": null,
"mp_size": 1,
"max_train_batch_size": null,
"min_train_batch_size": 1,
"max_train_micro_batch_size_per_gpu": 1.024000e+03,
"min_train_micro_batch_size_per_gpu": 1,
"num_tuning_micro_batch_sizes": 3
}
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] bfloat16_enabled ............. False
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] checkpoint_parallel_write_pipeline False
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] checkpoint_tag_validation_enabled True
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] checkpoint_tag_validation_fail False
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] comms_config ................. <deepspeed.comm.config.DeepSpeedCommsConfig object at 0x7fcf729e5fa0>
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] communication_data_type ...... None
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] compression_config ........... {'weight_quantization': {'shared_parameters': {'enabled': False, 'quantizer_kernel': False, 'schedule_offset': 0, 'quantize_groups': 1, 'quantize_verbose': False, 'quantization_type': 'symmetric', 'quantize_weight_in_forward': False, 'rounding': 'nearest', 'fp16_mixed_quantize': False, 'quantize_change_ratio': 0.001}, 'different_groups': {}}, 'activation_quantization': {'shared_parameters': {'enabled': False, 'quantization_type': 'symmetric', 'range_calibration': 'dynamic', 'schedule_offset': 1000}, 'different_groups': {}}, 'sparse_pruning': {'shared_parameters': {'enabled': False, 'method': 'l1', 'schedule_offset': 1000}, 'different_groups': {}}, 'row_pruning': {'shared_parameters': {'enabled': False, 'method': 'l1', 'schedule_offset': 1000}, 'different_groups': {}}, 'head_pruning': {'shared_parameters': {'enabled': False, 'method': 'topk', 'schedule_offset': 1000}, 'different_groups': {}}, 'channel_pruning': {'shared_parameters': {'enabled': False, 'method': 'l1', 'schedule_offset': 1000}, 'different_groups': {}}, 'layer_reduction': {'enabled': False}}
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] curriculum_enabled_legacy .... False
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] curriculum_params_legacy ..... False
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] data_efficiency_config ....... {'enabled': False, 'seed': 1234, 'data_sampling': {'enabled': False, 'num_epochs': 1000, 'num_workers': 0, 'curriculum_learning': {'enabled': False}}, 'data_routing': {'enabled': False, 'random_ltd': {'enabled': False, 'layer_token_lr_schedule': {'enabled': False}}}}
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] data_efficiency_enabled ...... False
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] dataloader_drop_last ......... False
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] disable_allgather ............ False
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] dump_state ................... False
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] dynamic_loss_scale_args ...... None
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] eigenvalue_enabled ........... False
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] eigenvalue_gas_boundary_resolution 1
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] eigenvalue_layer_name ........ bert.encoder.layer
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] eigenvalue_layer_num ......... 0
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] eigenvalue_max_iter .......... 100
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] eigenvalue_stability ......... 1e-06
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] eigenvalue_tol ............... 0.01
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] eigenvalue_verbose ........... False
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] elasticity_enabled ........... False
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] flops_profiler_config ........ {
"enabled": false,
"recompute_fwd_factor": 0.0,
"profile_step": 1,
"module_depth": -1,
"top_modules": 1,
"detailed": true,
"output_file": null
}
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] fp16_auto_cast ............... False
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] fp16_enabled ................. True
[2024-01-09 18:51:55,779] [INFO] [config.py:988:print] fp16_master_weights_and_gradients False
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] global_rank .................. 0
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] grad_accum_dtype ............. None
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] gradient_accumulation_steps .. 10
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] gradient_clipping ............ 1.0
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] gradient_predivide_factor .... 1.0
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] graph_harvesting ............. False
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] hybrid_engine ................ enabled=False max_out_tokens=512 inference_tp_size=1 release_inference_cache=False pin_parameters=True tp_gather_partition_size=8
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] initial_dynamic_scale ........ 65536
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] load_universal_checkpoint .... False
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] loss_scale ................... 0
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] memory_breakdown ............. False
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] mics_hierarchial_params_gather False
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] mics_shard_size .............. -1
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] monitor_config ............... tensorboard=TensorBoardConfig(enabled=False, output_path='', job_name='DeepSpeedJobName') wandb=WandbConfig(enabled=False, group=None, team=None, project='deepspeed') csv_monitor=CSVConfig(enabled=False, output_path='', job_name='DeepSpeedJobName') enabled=False
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] nebula_config ................ {
"enabled": false,
"persistent_storage_path": null,
"persistent_time_interval": 100,
"num_of_version_in_retention": 2,
"enable_nebula_load": true,
"load_path": null
}
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] optimizer_legacy_fusion ...... False
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] optimizer_name ............... adamw
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] optimizer_params ............. {'lr': 0.0003, 'weight_decay': 0.0, 'betas': (0.9, 0.95)}
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] pipeline ..................... {'stages': 'auto', 'partition': 'best', 'seed_layers': False, 'activation_checkpoint_interval': 0, 'pipe_partitioned': True, 'grad_partitioned': True}
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] pld_enabled .................. False
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] pld_params ................... False
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] prescale_gradients ........... False
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] scheduler_name ............... WarmupDecayLR
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] scheduler_params ............. {'total_num_steps': 5000, 'warmup_min_lr': 0, 'warmup_max_lr': 0.0003, 'warmup_num_steps': 100, 'warmup_type': 'linear'}
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] seq_parallel_communication_data_type torch.float32
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] sparse_attention ............. None
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] sparse_gradients_enabled ..... False
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] steps_per_print .............. 10
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] train_batch_size ............. 20
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] train_micro_batch_size_per_gpu 2
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] use_data_before_expert_parallel_ False
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] use_node_local_storage ....... False
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] wall_clock_breakdown ......... False
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] weight_quantization_config ... None
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] world_size ................... 1
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] zero_allow_untested_optimizer False
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] zero_config .................. stage=2 contiguous_gradients=True reduce_scatter=True reduce_bucket_size=500000000 use_multi_rank_bucket_allreduce=True allgather_partitions=True allgather_bucket_size=500000000 overlap_comm=True load_from_fp32_weights=True elastic_checkpoint=False offload_param=None offload_optimizer=None sub_group_size=1,000,000,000 cpu_offload_param=None cpu_offload_use_pin_memory=None cpu_offload=None prefetch_bucket_size=50,000,000 param_persistence_threshold=100,000 model_persistence_threshold=sys.maxsize max_live_parameters=1,000,000,000 max_reuse_distance=1,000,000,000 gather_16bit_weights_on_model_save=False stage3_gather_fp16_weights_on_model_save=False ignore_unused_parameters=True legacy_stage1=False round_robin_gradients=False zero_hpz_partition_size=1 zero_quantized_weights=False zero_quantized_nontrainable_weights=False zero_quantized_gradients=False mics_shard_size=-1 mics_hierarchical_params_gather=False memory_efficient_linear=True pipeline_loading_checkpoint=False override_module_apply=True
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] zero_enabled ................. True
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] zero_force_ds_cpu_optimizer .. True
[2024-01-09 18:51:55,780] [INFO] [config.py:988:print] zero_optimization_stage ...... 2
[2024-01-09 18:51:55,781] [INFO] [config.py:974:print_user_config] json = {
"train_micro_batch_size_per_gpu": 2,
"gradient_accumulation_steps": 10,
"optimizer": {
"type": "AdamW",
"params": {
"lr": 0.0003,
"weight_decay": 0.0,
"betas": [0.9, 0.95]
}
},
"scheduler": {
"type": "WarmupDecayLR",
"params": {
"total_num_steps": 5.000000e+03,
"warmup_min_lr": 0,
"warmup_max_lr": 0.0003,
"warmup_num_steps": 100,
"warmup_type": "linear"
}
},
"fp16": {
"enabled": true
},
"bf16": {
"enabled": false
},
"gradient_clipping": 1.0,
"zero_optimization": {
"stage": 2,
"contiguous_gradients": true,
"overlap_comm": true,
"reduce_scatter": true,
"reduce_bucket_size": 5.000000e+08,
"allgather_bucket_size": 5.000000e+08
}
}
<torch.utils.data.dataloader.DataLoader object at 0x7fd109aabca0>
0%| | 0/200 [00:00<?, ?it/s]Token indices sequence length is longer than the specified maximum sequence length for this model (757 > 512). Running this sequence through the model will result in indexing errors
100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 200/200 [01:18<00:00, 2.56it/s]
giou: 0.0759, ciou: 0.0648
same issue here
any update?
I meet the same issue，anybody fixed it？