| Commit message (Collapse) | Author | Age |
|---|
| |
|
|
| |
PiperOrigin-RevId: 211487989
|
| |
|
|
|
|
|
|
|
|
| |
conversions.
That is, instances of sp.ToString() are replaced with std::string(sp).
This will allow tensorflow::StringPiece::ToString to be removed, which is necessary before it can be replaced with absl::string_view.
PiperOrigin-RevId: 195162126
|
| |
|
|
| |
PiperOrigin-RevId: 157483181
|
| |
|
|
|
|
|
|
|
|
|
| |
This is motivated by, among other goals, the need to enhance memory efficiency during TFDBG's stepper operations. The stepper caches TensorHandles to already-continued-to tensors and use them as feeds if later continue-to actions depend on the tensors as transitive inputs. However, previously the TensorHandles had to be converted to Numpy arrays by calling eval() and the Numpy arrays were then fed back to next Session.run() calls. This mode of operation involved at least two unnecessary tensor-numpy and numpy-tensor copying.
This CL makes it possible to use the ResourceHandle representations TensorHandles directly as feed values, eliminating the need for the aforementioned copying.
To this end, the following changes are made
1) the underlying representations of TensorHandles are changed from string to ResourceHandle. A custom numpy struct type is created to allow ResourceHandle of the TensorHandle subtype to be fed during Session.run() calls.
2) added GetSessionHandleOpV2, which deprecates GetSessionHandleOp. The V2 op outputs a DT_RESOURCE Tensor, instead of a string Tensor in the deprecated version.
Change: 149672538
|
| |
|
|
| |
Change: 123900938
|
|
|
users. It allows a tensor produced by a run call to stay "in-place" so that a future run call can use it in-place. To achieve this, a run call can now return a handle of a tensor to the client, which can then be fed to a subsequent run call. This feature is complimentary to partial run, though there are some overlaps.
Here are a few properties of the current implementation:
1. Tensors are stored in the state of a session. The tensors are garbage collected if the client doesn't have a reference to the tensor or the session is closed.
2. There is no change to the current session API. We introduced two ops to manage the conversions between tensors and its handles. (There is a third op to garbage collect a tensor.) See the example below.
3. It fits quite well into the current feed-fetch design/implementation. It tries to reuse the graph (and caches) as much as possible so to make things efficient.
Below is a simple example. More examples can be found in sessopn_ops_test.py.
# Return a handle.
a = tf.constant(10)
b = tf.constant(5)
c = tf.mul(a, b)
h = tf.get_session_handle(c).eval()
# Feed a tensor handle.
f, x = tf.get_session_tensor(dtypes.int32)
y = tf.mul(x, 10)
result = sess.run(y, feed_dict={f: h.handle})
# result == 500
Change: 119481352
|
A. Unique TensorFlower
Peter Hawkins
Shanqing Cai
A. Unique TensorFlower
Yuan Yu