erryv3 / config.yaml
lapp0's picture
Upload Erry v3 checkpoint
e939777 verified
Raw History Blame Contribute Delete
5.38 kB
_wandb:
value:
cli_version: 0.28.1
e:
ciq56l2bkopdwoxv0hflf78jtgguipif:
codePath: wm_aes/models/maxgen/distillation/train_ltdec_fd.py
codePathLocal: wm_aes/models/maxgen/distillation/train_ltdec_fd.py
cpu_count: 64
cpu_count_logical: 128
cudaVersion: "12.4"
disk:
/:
total: "1064377483264"
used: "296780533760"
docker: .io/pytorch/pytorch@sha256:32df3983bd9e3ce902afea25d92ba07b9823f6edd5f6bcd4338cbdce256947e0
email: jxiong21029@gmail.com
executable: /usr/bin/python3
git:
commit: 1cedacc700d5773c99db8271d195488eb8fef8d4
remote: https://github.com/jxiong21029/wm-aes.git
gpu: NVIDIA H200
gpu_count: 8
gpu_nvidia:
- architecture: Hopper
cudaCores: 16896
memoryTotal: "150754820096"
name: NVIDIA H200
uuid: GPU-076fa98e-3cb2-3843-fe75-13162c377baa
- architecture: Hopper
cudaCores: 16896
memoryTotal: "150754820096"
name: NVIDIA H200
uuid: GPU-f1d22a92-4d2a-c9c7-cd96-1599bca5259b
- architecture: Hopper
cudaCores: 16896
memoryTotal: "150754820096"
name: NVIDIA H200
uuid: GPU-e05b1d37-290e-cf53-178a-9a8652cf9eae
- architecture: Hopper
cudaCores: 16896
memoryTotal: "150754820096"
name: NVIDIA H200
uuid: GPU-c1b1a344-b4f6-c878-c90d-0503be57bf49
- architecture: Hopper
cudaCores: 16896
memoryTotal: "150754820096"
name: NVIDIA H200
uuid: GPU-5591673c-dbd9-9f86-cc02-e2d69ec830ad
- architecture: Hopper
cudaCores: 16896
memoryTotal: "150754820096"
name: NVIDIA H200
uuid: GPU-e28338dd-4bde-6465-fef5-9652da2d3e68
- architecture: Hopper
cudaCores: 16896
memoryTotal: "150754820096"
name: NVIDIA H200
uuid: GPU-8db32a16-1099-aceb-dd5d-5e2e2f643c92
- architecture: Hopper
cudaCores: 16896
memoryTotal: "150754820096"
name: NVIDIA H200
uuid: GPU-52b92b01-6efe-7b7e-0f3e-6d288faa761a
host: sky-ceba-jerryxiong-d19f003d-head
memory:
total: "1690983649280"
os: Linux-5.15.0-142-generic-x86_64-with-glibc2.39
program: /app/wm_aes/models/maxgen/distillation/train_ltdec_fd.py
python: CPython 3.12.3
root: /mnt/data/wm-aes
startedAt: "2026-07-30T04:13:08.033082Z"
writerId: ciq56l2bkopdwoxv0hflf78jtgguipif
m: []
python_version: 3.12.3
t:
"1":
- 1
- 41
"2":
- 1
- 41
"3":
- 16
- 61
"4": 3.12.3
"5": 0.28.1
"12": 0.28.1
"13": linux-x86_64
base_ckpt:
value: /mnt/data/wm-aes/wandb/run-20260723_060955-6j0fhgdf
base_lr:
value: 0.0001
base_wd:
value: 1e-05
betas:
value:
- 0.9
- 0.99
compile_mode:
value: default
dtype:
value: bfloat16
early_valid_steps:
value:
- 1
- 100
- 500
fd_cov_eps:
value: 1e-06
fd_decay:
value: 0.99
feature_fd_weight:
value: 0.1
input_sizes:
value:
- - 720
- 1280
- - 288
- 640
lora_lr:
value: 1e-05
lr_cooldown_ratio:
value: 0
lr_cooldown_start:
value: 5000
mse_weight:
value: 1
muon_exclusions:
value:
- dwconv_weight
- latent_proj.weight
- out_proj.weight
n_steps:
value: 10000
n_video_seq_per_device:
value:
- 1
- 1
n_visuals:
value: 16
num_loader_workers:
value: 4
profile:
value: false
profile_dir:
value: logs/profile/
profile_record_shapes:
value: false
profile_with_stack:
value: false
run_base_dir:
value: /mnt/data/wm-aes/
run_name:
value: null
scalar_lr:
value: 1e-05
seed:
value: 0
seq_len:
value: 80
tqdm_kwargs:
value:
mininterval: 3
ncols: 100
smoothing: 1
train_vid_indexes:
value:
- wm_aes/data/indexes/owlc_train_10000.jsonl
- wm_aes/data/indexes/ov1m_train_100000.jsonl
use_wandb:
value: true
valid_interval:
value: 1000
valid_size:
value: 1024
valid_vid_indexes:
value:
- wm_aes/data/indexes/owlc_eval_100.jsonl
- wm_aes/data/indexes/ov1m_eval_1000.jsonl
wandb_group:
value: dito-v1
wd_cooldown:
value: false
wd_exclusions:
value:
- dwconv_weight