Mkulakow/mediapipe kfs rest test #2377
Conversation
michalkulakowski
force-pushed
the
mkulakow/mediapipe_kfs_rest_test
branch
2 times, most recently
from
84cb674
to
c1c3491
Compare
src/test/kfs_rest_test.cpp
Outdated
| } | ||
| } | ||
|
|
||
| static void testInferenceNegative(int headerLength, std::string& request_body, std::unique_ptr<HttpRestApiHandler>& handler, ovms::Status status) { |
There was a problem hiding this comment.
| static void testInferenceNegative(int headerLength, std::string& request_body, std::unique_ptr<HttpRestApiHandler>& handler, ovms::Status status) { | |
| static void testInferenceNegative(int headerLength, std::string& request_body, std::unique_ptr<HttpRestApiHandler>& handler, ovms::Status processorStatus) { |
It's not worth at this point probably to add possibility for testing parsing components, but eventually we could add it later here as well.
| @@ -0,0 +1,31 @@ | |||
| # | |||
| # Copyright 2023 Intel Corporation | |||
There was a problem hiding this comment.
| # Copyright 2023 Intel Corporation | |
| # Copyright 2024 Intel Corporation |
| class HttpRestApiHandlerWithMediapipeForkTest : public HttpRestApiHandlerWithMediapipe { | ||
| public: | ||
| void SetUp() { | ||
| SetUpServer("/ovms/src/test/mediapipe/config_summator.json"); |
There was a problem hiding this comment.
Why we remove this json & pbtxt?
michalkulakowski
force-pushed
the
mkulakow/mediapipe_kfs_rest_test
branch
from
c1c3491
to
ddefb38
Compare
| if (precision == ov::element::Type_t::string) { | ||
| expectedBytes = 0; | ||
| for (auto contents : request.inputs(inputIndex).contents().bytes_contents()) { | ||
| expectedBytes = expectedBytes + contents.size() + sizeof(uint32_t); |
There was a problem hiding this comment.
Isn't size written in first bytes of the input? If so I suggest keeping the order also here.
| expectedBytes = expectedBytes + contents.size() + sizeof(uint32_t); | |
| expectedBytes += sizeof(uint32_t) + contents.size(); |
| case ov::element::Type_t::string: { | ||
| uint32_t offset = 0; | ||
| for (auto contents : request.inputs(inputIndex).contents().bytes_contents()) { | ||
| uint32_t size = contents.size(); |
There was a problem hiding this comment.
This is used only below as a buffer to copy from. Do we need that? If so, can we make it const and replace following contents.size() calls?
src/python/ovms_py_tensor.hpp
Outdated
| @@ -83,6 +85,7 @@ const std::unordered_map<std::string, py::ssize_t> bufferFormatToItemsize{ | |||
| {"e", 2}, | |||
| {"f", 4}, | |||
| {"d", 8}, | |||
| {"s", 1}, | |||
There was a problem hiding this comment.
How is it 1? BYTES type was considered custom type since in a way it's defined in KServe, it's impossible to determine size of a single element.
src/test/kfs_rest_test.cpp
Outdated
| #pragma GCC diagnostic push | ||
| #pragma GCC diagnostic ignored "-Wnarrowing" | ||
|
|
||
| TEST_F(HttpRestApiHandlerWithMediapipeForkTest, inferRequestFP32) { | ||
| TEST_F(HttpRestApiHandlerWithMediapipe, inferRequestFP64) { |
There was a problem hiding this comment.
There are a lot of very similar tests here. Maybe we could parametrize some of them? The only differences I see are in datatype and status.
https://jira.devtools.intel.com/browse/CVS-134120
https://jira.devtools.intel.com/browse/CVS-134803