Default use_cache=False in eval harness integration (#774)

* Set `use_cache=False` in run_eval_harness * Update NeoXArgs docs automatically * Update NeoXArgs docs automatically --------- Co-authored-by: github-actions <[email protected]> Co-authored-by: Quentin Anthony <[email protected]>
EleutherAI · Feb 3, 2023 · 26ef16d · 26ef16d
1 parent 0c831c3
commit 26ef16d
Show file tree

Hide file tree

Showing 2 changed files with 11 additions and 1 deletion.
diff --git a/configs/neox_arguments.md b/configs/neox_arguments.md
@@ -111,7 +111,7 @@ Logging Arguments
 
 - **git_hash**: str
 
- Default = 3463ca4
+ Default = 7ed7bc6
 
  current git hash of repository
 
@@ -1018,6 +1018,15 @@ Training Arguments
 
 
 
+- **use_shared_fs**: bool
+
+ Default = True
+
+ Whether to use a shared filesystem for data loading. If False, local rank 0 on all nodes will preprocess the data,
+ otherwise only global rank 0 will preprocess the data. This is implemented in megatron/data/gpt2_dataset.py::_build_index_mappings.
+
+
+
 - **train_data_paths**: list
 
  Default = None

diff --git a/eval_tasks/eval_adapter.py b/eval_tasks/eval_adapter.py
@@ -462,4 +462,5 @@ def run_eval_harness(
  eval_tasks=eval_tasks,
  num_fewshot=num_fewshot,
  bootstrap_iters=bootstrap_iters,
+ use_cache=False,
  )