Don't actually add a duplicate

pytorch · Jul 28, 2023 · 4f19a13 · 4f19a13
1 parent 5262644
commit 4f19a13
Show file tree

Hide file tree

Showing 5 changed files with 4 additions and 38 deletions.
diff --git a/torchbenchmark/models/llama_v2_7b_8h/__init__.py b/torchbenchmark/models/llama_v2_7b_8h/__init__.py
diff --git a/torchbenchmark/models/llama_v2_7b_8h/install.py b/torchbenchmark/models/llama_v2_7b_8h/install.py
diff --git a/torchbenchmark/models/llama_v2_7b_8h/metadata.yaml b/torchbenchmark/models/llama_v2_7b_8h/metadata.yaml
diff --git a/torchbenchmark/util/framework/huggingface/model_factory.py b/torchbenchmark/util/framework/huggingface/model_factory.py
@@ -30,8 +30,6 @@
     'hf_Whisper': (1024, 1024, 'WhisperConfig()', 'AutoModelForAudioClassification'),
     # default num_hidden_layers=32 but that OOMs, feel free to change this config to something more real
     'llama_v2_7b_16h' : (512,512, 'LlamaConfig(num_hidden_layers=16)', 'AutoModelForCausalLM'),
-    # an even smaller model since 16h OOMs for our optimizer benchmarks
-    'llama_v2_7b_8h' : (512,512, 'LlamaConfig(num_hidden_layers=8)', 'AutoModelForCausalLM'),
 }
 
 cpu_input_slice = {

diff --git a/userbenchmark/optim/run.py b/userbenchmark/optim/run.py
@@ -249,6 +249,10 @@ def get_unstable_models() -> Set[str]:
     # Skip models deemed unstable by torch-nightly
     {'model': m} for m in unstable_models
 ] + [
+    # 16h currently OOMs, but once it supports train, we should remove this line
+    # See tracker https://github.com/pytorch/benchmark/issues/1793
+    {'model': 'llama_v2_7b_16h'}
+] +[
     # SparseAdam does not support dense gradients
     {'optim': 'SparseAdam', 'model': m} for m in DENSE_MODELS
 ] + [