PTQ-Level Efficiency, QAT-Level Accuracy | 训练后量化的效率,量化感知训练的精度
update shells (#112)
WeiweiZhang13c214db0f126
diff --git a/examples/language-modeling/run_autoround.sh b/examples/language-modeling/run_autoround.shindex ddf13aa1a..38a907f55 100644--- a/examples/language-modeling/run_autoround.sh+++ b/examples/language-modeling/run_autoround.sh@@ -8,11 +8,10 @@ python3 main.py \--model_name $model_name \--bits 4 \--group_size 128 \---enable_minmax_tuning \---amp \--iters 200 \--deployment_device 'fake,cpu' \--scale_dtype 'fp32' \--eval_bs 32 \--output_dir "./tmp_autoround"+diff --git a/examples/language-modeling/run_autoround_on_gaudi.sh b/examples/language-modeling/run_autoround_on_gaudi.shindex ab862dc0c..2dd0f4b62 100644--- a/examples/language-modeling/run_autoround_on_gaudi.sh+++ b/examples/language-modeling/run_autoround_on_gaudi.sh@@ -1,15 +1,12 @@#!/bin/bashset -xmodel_name="Intel/neural-chat-7b-v3-3"-eval_bs=1python3 main.py \- --amp \--model_name $model_name \- --device "hpu" \--group_size 128 \--bits 4 \- --enable_minmax_tuning \- --eval_bs $eval_bs \+ --deployment_device "fake" \--output_dir "./tmp_autoround"+diff --git a/examples/language-modeling/run_xpu.sh b/examples/language-modeling/run_xpu.shindex dd651b8d6..26286f9c3 100644--- a/examples/language-modeling/run_xpu.sh+++ b/examples/language-modeling/run_xpu.sh@@ -1,22 +1,16 @@#!/bin/bash-+set -xdevice=1-eval_bs=32-model_name="/facebook/opt-125m"-+model_name="facebook/opt-125m"CUDA_VISIBLE_DEVICES=$device \python3 main.py \--model_name $model_name \---device 0 \---group_size 128 \--bits 4 \---iters 10 \---deployment_device 'xpu,fake' \---quantize_layers_outside_blocks \+--group_size 32 \+--deployment_device "xpu,fake" \--disable_low_gpu_mem_usage \--disable_eval \---output_dir "./test-export" \--scale_dtype "fp16" \---group_size 32 \-2>&1 | tee -a 4.16_xpu.log+--output_dir "./tmp_autoround" \+