fix enwik8 errors

EleutherAI · Jan 5, 2021 · a38537e · a38537e
1 parent ee7aad0
commit a38537e
Show file tree

Hide file tree

Showing 3 changed files with 4 additions and 4 deletions.
diff --git a/configs/base_model.json b/configs/base_model.json
@@ -1,7 +1,7 @@
 {
  "dataset": {
  "name": "enwik8",
- "path": "./data/enwik8.tar.gz"
+ "path": "./data/enwik8.gz"
  },
  "num_epochs": 10,
  "vocab_size": 256,

diff --git a/gpt_neox/__init__.py b/gpt_neox/__init__.py
@@ -1,5 +1,5 @@
 from gpt_neox.autoregressive_wrapper import AutoregressiveWrapper
-from gpt_neox.data_utils import get_tokenizer
+from gpt_neox.data_utils import get_tokenizer, read_enwik8_data
 from gpt_neox.datasets import TextSamplerDataset, GPT2Dataset
 from gpt_neox.gpt_neox import GPTNeoX
 from gpt_neox.utils import *

diff --git a/train_enwik8.py b/train_enwik8.py
@@ -8,7 +8,7 @@
 from torch.utils.data import DataLoader
 from tqdm.auto import trange
 
-from gpt_neox import (GPTNeoX, AutoregressiveWrapper, TextSamplerDataset, download_dataset,
+from gpt_neox import (GPTNeoX, AutoregressiveWrapper, TextSamplerDataset,
  cycle, prepare_optimizer_parameters, decode_tokens, read_enwik8_data, is_main, prepare_data)
 
 
@@ -54,7 +54,7 @@ def get_params(model):
  torch.distributed.barrier()
 
 # prepare enwik8 data
-data_train, data_val = read_enwik8_data(dset_params["data_path"])
+data_train, data_val = read_enwik8_data(dset_params["path"])
 train_dataset = TextSamplerDataset(data_train, params["seq_len"])
 val_dataset = TextSamplerDataset(data_val, params["seq_len"])
 val_loader = cycle(DataLoader(val_dataset, batch_size=params["batch_size"]))