open_flamingo icon indicating copy to clipboard operation
open_flamingo copied to clipboard

'media_token_id' AttributeError

Open olo126 opened this issue 2 years ago • 0 comments

I set up an environment following the installation instructions and am trying to run train.py with the default lm, opt-1.3b,

torchrun --nnodes=1 --nproc_per_node=2 train/train.py \
  --lm_path facebook/opt-1.3b \
  --tokenizer_path facebook/opt-30b \
  --model_family flamingo \
  --cross_attn_every_n_layers 4 \
  --dataset_resampled \
  --batch_size_laion 32 \
  --train_num_samples_laion 250000 \
  --loss_multiplier_laion 0.2 \
  --workers=2 \
  --run_name "test" \
  --num_epochs 100 \
  --laion_shards "/mmfs1/gscratch/efml/anasa2/eval_benchmark/coco_wds/{000..064}_{0000..00012}.tar" \
  --report_to_wandb

But I get the following error:

Traceback (most recent call last): File "/mmfs1/gscratch/efml/olo126/open_flamingo/open_flamingo/train/train.py", line 450, in main() File "/mmfs1/gscratch/efml/olo126/open_flamingo/open_flamingo/train/train.py", line 430, in main train_one_epoch( File "/mmfs1/gscratch/efml/olo126/open_flamingo/open_flamingo/train/train_utils.py", line 81, in train_one_epoch (input_ids == model.media_token_id).sum().item() File "/mmfs1/gscratch/xlab/olo126/miniconda3/envs/openflamingo/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1614, in getattr raise AttributeError("'{}' object has no attribute '{}'".format( AttributeError: 'DistributedDataParallel' object has no attribute 'media_token_id'

olo126 avatar Nov 09 '23 20:11 olo126