From 2952d4cef2d8ea27d6205f6f8fc8309238532983 Mon Sep 17 00:00:00 2001
From: whs <wanghaoshuang@baidu.com>
Date: Tue, 7 Jun 2022 17:16:26 +0800
Subject: [PATCH] Skip auto tuning when shape of model's  input is variable
 (#1155)

---
 .pre-commit-config.yaml                      |  1 +
 paddleslim/analysis/extract_features.py      |  1 -
 paddleslim/auto_compression/auto_strategy.py | 12 +++++++++-
 paddleslim/auto_compression/utils/predict.py | 24 ++++++++++++++++++++
 4 files changed, 36 insertions(+), 2 deletions(-)

diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
index 9d849f12..6bc37d8b 100644
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -12,6 +12,7 @@ repos:
     sha: 0d79c0c469bab64f7229c9aca2b1186ef47f0e37
     hooks:
     -   id: yapf
+        language_version: python3.9
         files: (.*\.(py|bzl)|BUILD|.*\.BUILD|WORKSPACE)$
 -   repo: https://github.com/pre-commit/pre-commit-hooks
     sha: 5bf6c09bfa1297d3692cadd621ef95f1284e33c0
diff --git a/paddleslim/analysis/extract_features.py b/paddleslim/analysis/extract_features.py
index 17fe8822..a960c393 100644
--- a/paddleslim/analysis/extract_features.py
+++ b/paddleslim/analysis/extract_features.py
@@ -50,7 +50,6 @@ def get_features_from_paramkey(param_key, op_type, data_type):
     """Get op's parameters according to the key of latency table
     """
     features = None
-
     if 'conv2d' in op_type:
         if data_type == 'fp16':
             quant_bits = 'bit_length=16'
diff --git a/paddleslim/auto_compression/auto_strategy.py b/paddleslim/auto_compression/auto_strategy.py
index 8c1bc23a..d4820dfc 100644
--- a/paddleslim/auto_compression/auto_strategy.py
+++ b/paddleslim/auto_compression/auto_strategy.py
@@ -16,7 +16,7 @@ import os
 import logging
 import platform
 from ..common import get_logger
-from .utils.predict import predict_compressed_model
+from .utils.predict import predict_compressed_model, with_variable_shape
 from .strategy_config import *
 
 _logger = get_logger(__name__, level=logging.INFO)
@@ -151,6 +151,16 @@ def prepare_strategy(executor,
     """ prepare compression config automatically """
     final_strategy = None
 
+    if with_variable_shape(
+            model_dir,
+            model_filename=model_filename,
+            params_filename=params_filename):
+        deploy_hardware = None
+        _logger.warning(
+            "The model's inputs have variable shape. "
+            "And the latency predictor doesn't support variable shape. "
+            "So auto tuning will be skipped and a default strategy will be chosen."
+        )
     ### use hardware latency tabel if support
     if deploy_hardware is not None:
         compressed_time_dict = predict_compressed_model(
diff --git a/paddleslim/auto_compression/utils/predict.py b/paddleslim/auto_compression/utils/predict.py
index 2cf21fb9..2bc316c2 100644
--- a/paddleslim/auto_compression/utils/predict.py
+++ b/paddleslim/auto_compression/utils/predict.py
@@ -6,6 +6,30 @@ from .prune_model import get_sparse_model, get_prune_model
 from .fake_ptq import post_quant_fake
 
 
+def with_variable_shape(model_dir, model_filename=None, params_filename=None):
+    """
+    Whether the shape of model's input is variable.
+    Args:
+        path_prefix(str | None): Directory path to save model + model name without suffix.
+        model_filename(str): specify model_filename if you don't want to use default name. Default : 'None'.
+        params_filename(str): specify params_filename if you don't want to use default name. Default : 'None'.
+    Returns:
+        bool: Whether the shape of model's input is variable.
+    """
+    paddle.enable_static()
+    exe = paddle.static.Executor(paddle.CPUPlace())
+    [inference_program, feed_target_names, fetch_targets] = (
+        paddle.static.load_inference_model(
+            model_dir,
+            exe,
+            model_filename=model_filename,
+            params_filename=params_filename))
+    for var_ in inference_program.list_vars():
+        if var_.name in feed_target_names:
+            if var_.shape.count(-1) > 1:
+                return True
+
+
 def predict_compressed_model(executor,
                              places,
                              model_dir,
-- 
GitLab