Merge pull request #2712 from AI-Hypercomputer:shuningjin-fix

PiperOrigin-RevId: 834025121

Author: Google-ML-Automation

end_to_end/tpu/gpt_oss/20b/test_gpt_oss.sh

@@ -49,6 +49,8 @@ export UNSCANNED_CKPT_PATH=${BASE_OUTPUT_PATH}/unscanned/0/items
 # Non-Googlers please remember to point `DATASET_PATH` to the GCS bucket where you have your training data
 export DATASET_PATH=gs://maxtext-dataset
 
+export LIBTPU_INIT_ARGS='--xla_tpu_scoped_vmem_limit_kib=81920'
+
 # Test whether the forward pass logits match the golden logits
 # default golden_logits_path=/deps/src/MaxText/test_assets/golden_data_{MODEL_NAME}.jsonl, copied from gs://maxtext-test-assets/golden_data_${MODEL_NAME}.jsonl
 python3 -m tests.forward_pass_logit_checker "${MAXTEXT_PKG_DIR:-${MAXTEXT_REPO_ROOT:-$PWD}/src/MaxText}/"configs/base.yml base_output_directory=${BASE_OUTPUT_PATH} run_name=forward_logits_check model_name=${MODEL_NAME} load_parameters_path=${UNSCANNED_CKPT_PATH} scan_layers=false attention=dot_product sparse_matmul=True megablox=True per_device_batch_size=1 max_target_length=4 max_prefill_predict_length=4 dtype=float32 --atol=0.1 --rtol=0.1 --max_kl_div=3e-4

src/MaxText/configs/types.py

@@ -61,6 +61,10 @@ class MatmulPrecision(str, Enum):
   DEFAULT = "default"
   HIGH = "high"
   HIGHEST = "highest"
+  # same as default
+  BFLOAT16 = "bfloat16"
+  # same as highest
+  FLOAT32 = "float32"
 
 
 class QuantizationType(str, Enum):