Merge pull request #11 from xixiaoyao/master

fix bugs

paddlepalm/mtl_controller.py

@@ -626,7 +626,8 @@ class Controller(object):
             rt_outputs = self.exe.run(train_program, fetch_list=fetch_list)
             rt_outputs = {k:v for k,v in zip(fetch_names, rt_outputs)}
             rt_task_id = np.squeeze(rt_outputs['__task_id']).tolist()
-            assert (not isinstance(rt_task_id, list)) or len(set(rt_task_id)) == 1, rt_task_id
+            # 注意注释掉这一行之后，训练日志实际是错误的
+            # assert (not isinstance(rt_task_id, list)) or len(set(rt_task_id)) == 1, rt_task_id
             rt_task_id = rt_task_id[0] if isinstance(rt_task_id, list) else rt_task_id
             cur_task = instances[rt_task_id]
 

paddlepalm/utils/reader_helper.py

@@ -72,7 +72,8 @@ def create_net_inputs(input_attrs, async=False, iterator_fn=None, dev_count=1, n
 
     if async:
         assert iterator_fn is not None, "iterator_fn is needed for building async input layer."
-        reader = fluid.io.PyReader(inputs, capacity=dev_count*n_prefetch, iterable=False)
+        # reader = fluid.io.PyReader(inputs, capacity=dev_count*n_prefetch, iterable=False)
+        reader = fluid.io.PyReader(inputs, capacity=dev_count, iterable=False)
         reader.decorate_batch_generator(iterator_fn)
         reader.start()
 
@@ -153,7 +154,7 @@ def create_joint_iterator_fn(iterators, iterator_prefixes, joint_shape_and_dtype
             
             for i in range(dev_count):
                 # results = _zero_batch(joint_shape_and_dtypes, batch_size=batch_size)
-                # results[0] = task_id_tensor
+                results[0] = task_id_tensor
                 if id in outbuf:
                     outputs = outbuf[id]
                     del outbuf[id]

run_demo1.sh

-export CUDA_VISIBLE_DEVICES=0,1,2,3
-export FLAGS_fraction_of_gpu_memory_to_use=0.1
-export FLAGS_eager_delete_tensor_gb=0
+export CUDA_VISIBLE_DEVICES=0
 
 python demo1.py
 

run_demo2.sh

-export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
-export FLAGS_fraction_of_gpu_memory_to_use=0.1
-export FLAGS_eager_delete_tensor_gb=0
+export CUDA_VISIBLE_DEVICES=0,1,2,3
 
 python demo2.py