add schema to control asym/sym (#1545)

intel · May 14, 2024 · 49ab653 · 49ab653
1 parent ae7a4ae
commit 49ab653
Showing 1 changed file with 8 additions and 4 deletions.
diff --git a/examples/huggingface/pytorch/text-generation/quantization/run_tuning.sh b/examples/huggingface/pytorch/text-generation/quantization/run_tuning.sh
@@ -20,6 +20,7 @@ function init_params {
   script="run_generation_sq.py"
   alpha=0.5
   weight_dtype="int4_clip"
+  scheme="asym"
   for var in "$@"
   do
     case $var in
@@ -47,6 +48,9 @@ function init_params {
        --bits=*)
            bits=$(echo $var |cut -f2 -d=)
        ;;
+       --scheme=*)
+           scheme=$(echo $var |cut -f2 -d=)
+       ;;
       *)
           echo "Error: No such parameter: ${var}"
           exit 1
@@ -248,31 +252,31 @@ function run_tuning {
         script="run_generation_sq.py"
     elif [ "${topology}" = "llama2_7b_gptq" ]; then
         model_name_or_path="meta-llama/Llama-2-7b-hf"
-        extra_cmd=$extra_cmd" --woq --bits ${bits} --compute_dtype fp32 --scheme asym --calib_iters 100"
+        extra_cmd=$extra_cmd" --woq --bits ${bits} --compute_dtype fp32 --scheme ${scheme} --calib_iters 100"
         extra_cmd=$extra_cmd" --woq_algo "GPTQ" --desc_act --blocksize 128 --max_input_length 2048 "
         extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
         extra_cmd=$extra_cmd" --trust_remote_code"
         extra_cmd=$extra_cmd" --weight_dtype ${weight_dtype}"
         script="run_generation_cpu_woq.py"
     elif [ "${topology}" = "mistral_7b_autoround" ]; then
         model_name_or_path="/tf_dataset2/models/pytorch/Mistral-7B-v0.1"
-        extra_cmd=$extra_cmd" --woq --bits ${bits} --compute_dtype fp32 --scheme asym "
+        extra_cmd=$extra_cmd" --woq --bits ${bits} --compute_dtype fp32 --scheme ${scheme} "
         extra_cmd=$extra_cmd" --woq_algo "AutoRound" --desc_act --group_size 128 --calib_len 2048 --calib_iters 100"
         extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
         extra_cmd=$extra_cmd" --trust_remote_code"
         extra_cmd=$extra_cmd" --weight_dtype ${weight_dtype}"
         script="run_generation_cpu_woq.py"
     elif [ "${topology}" = "mistral_7b_rtn" ]; then
         model_name_or_path="/tf_dataset2/models/pytorch/Mistral-7B-v0.1"
-        extra_cmd=$extra_cmd" --woq --bits ${bits} --compute_dtype fp32 --scheme asym "
+        extra_cmd=$extra_cmd" --woq --bits ${bits} --compute_dtype fp32 --scheme ${scheme} "
         extra_cmd=$extra_cmd" --woq_algo "Rtn" "
         extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
         extra_cmd=$extra_cmd" --trust_remote_code"
         extra_cmd=$extra_cmd" --weight_dtype ${weight_dtype}"
         script="run_generation_cpu_woq.py"
     elif [ "${topology}" = "mistral_7b_gptq" ]; then
         model_name_or_path="/tf_dataset2/models/pytorch/Mistral-7B-v0.1"
-        extra_cmd=$extra_cmd" --woq --bits ${bits} --compute_dtype fp32 --scheme asym --calib_iters 100"
+        extra_cmd=$extra_cmd" --woq --bits ${bits} --compute_dtype fp32 --scheme ${scheme} --calib_iters 100"
         extra_cmd=$extra_cmd" --woq_algo "GPTQ" --desc_act --blocksize 128 --max_input_length 2048 --group_size 128"
         extra_cmd=$extra_cmd" --output_dir ${tuned_checkpoint}"
         extra_cmd=$extra_cmd" --trust_remote_code"