update autoround example for release test

Signed-off-by: xin3he <[email protected]>
intel · Dec 13, 2024 · cca2c03 · cca2c03
1 parent bcae95b
commit cca2c03
Show file tree

Hide file tree

Showing 3 changed files with 15 additions and 0 deletions.
diff --git a/examples/.config/model_params_pytorch_3x.json b/examples/.config/model_params_pytorch_3x.json
@@ -129,6 +129,13 @@
       "main_script": "run_clm_no_trainer.py",
       "batch_size": 1
     },
+    "opt_125m_woq_autoround_int4_hpu":{
+      "model_src_dir": "nlp/huggingface_models/language-modeling/quantization/weight_only",
+      "dataset_location": "",
+      "input_model": "",
+      "main_script": "run_clm_no_trainer.py",
+      "batch_size": 1
+    },
     "opt_125m_woq_autotune_int4":{
       "model_src_dir": "nlp/huggingface_models/language-modeling/quantization/weight_only",
       "dataset_location": "",

diff --git a/...ytorch/nlp/huggingface_models/language-modeling/quantization/weight_only/run_benchmark.sh b/...ytorch/nlp/huggingface_models/language-modeling/quantization/weight_only/run_benchmark.sh
@@ -101,6 +101,9 @@ function run_benchmark {
     elif [ "${topology}" = "opt_125m_woq_autoround_int4" ]; then
         model_name_or_path="facebook/opt-125m"
         extra_cmd=$extra_cmd" --woq_algo AutoRound"
+    elif [ "${topology}" = "opt_125m_woq_autoround_int4_hpu" ]; then
+        model_name_or_path="facebook/opt-125m"
+        extra_cmd=$extra_cmd" --woq_algo AutoRound"
     elif [ "${topology}" = "opt_125m_woq_autotune_int4" ]; then
         model_name_or_path="facebook/opt-125m"
     fi

diff --git a/...pi/pytorch/nlp/huggingface_models/language-modeling/quantization/weight_only/run_quant.sh b/...pi/pytorch/nlp/huggingface_models/language-modeling/quantization/weight_only/run_quant.sh
@@ -95,6 +95,11 @@ function run_tuning {
     elif [ "${topology}" = "opt_125m_woq_autoround_int4" ]; then
         model_name_or_path="facebook/opt-125m"
         extra_cmd=$extra_cmd" --woq_algo AutoRound --woq_bits 4 --woq_group_size 128 --woq_scheme asym --autoround_iters 200 --autoround_nsamples 500"
+    elif [ "${topology}" = "opt_125m_woq_autoround_int4_hpu" ]; then
+        model_name_or_path="facebook/opt-125m"
+        extra_cmd=$extra_cmd" --woq_algo AutoRound --woq_bits 4 --woq_group_size 128 --woq_scheme asym --autoround_iters 200 --autoround_nsamples 500"
+        export PT_ENABLE_INT64_SUPPORT=1
+        export PT_HPU_LAZY_MODE=0
     elif [ "${topology}" = "opt_125m_woq_autotune_int4" ]; then
         model_name_or_path="facebook/opt-125m"
         extra_cmd=$extra_cmd" --woq_algo AutoTune --woq_bits 4"