gpt2_pp_2gpu | kastan Overview – Weights & Biases

Skip to main content

Kastan's group workspace

Group: gpt2_pp_2gpu

2

1-1

of 1

Tags

Notes

Author

State

Crashed

Start time

June 21st, 2022 7:09:12 PM

Runtime

17m 14s

Tracked hours

17m 16s

Run path

kastan/col_ai/248q61tv

OS

Linux-4.18.0-348.23.1.el8_5.x86_64-x86_64-with-glibc2.10

Python version

3.8.12

Git repository

git clone https://github.com/hpcaitech/ColossalAI-Examples.git

Git state

git checkout -b "gpt2_pp_2gpu" 1371a5ba8bc454d081ecb0d88f7b3f801dde01f6

Command

train_gpt.py --config gpt2_configs/gpt2_pp_2gpu.py --from_torch

System Hardware

CPU count	128
GPU count	4
GPU type	NVIDIA A100-SXM4-40GB

W&B CLI Version

0.12.18

Group

Config parameters are your model's inputs. Learn more

▶
Config parameters:{} 22 keys
- backend:
  "nccl"
- BATCH_SIZE:
  8
- colossal_config_file:
  "gpt2_configs/gpt2_pp_2gpu.py"
- config:
  "gpt2_configs/gpt2_pp_2gpu.py"
- data_dir:
  "/u/kastan/colossal/data/train_data_FINAL.json"
- ▶
  fp16:{} 1 key
  - mode:
    "AMP_TYPE.NAIVE"
- from_torch:
  true
- gpt2_small:
  "titans.model.gpt.gpt.gpt2_small"
- HIDDEN_SIZE:
  768
- host:
  null
- local_rank:
  null
- ▶
  loss:{} 1 key
  - type:
    "titans.loss.lm_loss.gpt_lmloss.GPTLMLoss"
- ▶
  model:{} 2 keys
  - checkpoint:
    true
  - type:
    "titans.model.gpt.gpt.gpt2_small"
- NUM_EPOCHS:
  60
- NUM_MICRO_BATCHES:
  4
- ▶
  optimizer:{} 3 keys
  - lr:
    0.00015
  - type:
    "torch.optim.adam.Adam"
  - weight_decay:
    0.01
- ▶
  parallel:{} 2 keys
  - pipeline:
    2
  - ▶
    tensor:{} 2 keys
    - mode:
      null
    - size:
      1
- PIPELINE:
  2
- port:
  null
- rank:
  null
- SEQ_LEN:
  1,024
- world_size:
  null

Summary metrics are your model's outputs. Learn more

▶
Summary metrics:{} 5 keys
- hello_message:
  "Hi, we got started"
- loss:
  16.113948822021484
- per_step_metrics.loss:
  16.113948822021484
- per_step_metrics.samples_per_sec:
  1.7712
- per_step_metrics.Tflops:
  2.1421

This run produced these artifacts as outputs. Total: 1. Learn more

run-248q61tv-history:v0

Loading...