Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[api] Let single handle flexible filter #574

Open
wants to merge 3 commits into
base: main
Choose a base branch
from
Open

[api] Let single handle flexible filter #574

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
120c938
[api] Add a new nnfw `EXECUTORCH_LLAMA`
anyj0527 Oct 29, 2024
7485a33
[api] Let single handle flexible filter
anyj0527 Oct 29, 2024
ed9a143
[test] Add a disabled test for executorch-llama
anyj0527 Oct 29, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions c/include/ml-api-common.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -75,6 +75,7 @@ typedef enum {
ML_NNFW_TYPE_NCNN = 18, /**< Tencent ncnn (Since 9.0) */
ML_NNFW_TYPE_TENSORRT = 19, /**< NVidia Tensor-RT (Since 9.0) */
ML_NNFW_TYPE_QNN = 20, /**< Qualcomm QNN (Qualcomm® AI Engine Direct) (Since 9.0) */
ML_NNFW_TYPE_EXECUTORCH_LLAMA = 21, /**< ExecuTorch Llama runner */
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Add Tizen version requirement

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you separate new PR and add github label for next ACR?

ML_NNFW_TYPE_SNAP = 0x2001, /**< SNAP (Samsung Neural Acceleration Platform), only for Android. (Since 6.0) */
} ml_nnfw_type_e;

Expand Down
19 changes: 19 additions & 0 deletions c/src/ml-api-inference-single.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -112,6 +112,7 @@ static const char *ml_nnfw_subplugin_name[] = {
[ML_NNFW_TYPE_NCNN] = "ncnn",
[ML_NNFW_TYPE_TENSORRT] = "tensorrt",
[ML_NNFW_TYPE_QNN] = "qnn",
[ML_NNFW_TYPE_EXECUTORCH_LLAMA] = "executorch-llama",
NULL
};

Expand All @@ -137,6 +138,7 @@ typedef struct
gboolean invoking; /**< invoke running flag */
ml_tensors_data_h in_tensors; /**< input tensor wrapper for processing */
ml_tensors_data_h out_tensors; /**< output tensor wrapper for processing */
gboolean is_flexible; /**< true if tensor filter handles flexible input/output */

GList *destroy_data_list; /**< data to be freed by filter */
} ml_single;
Expand Down Expand Up @@ -778,6 +780,11 @@ ml_single_set_info_in_handle (ml_single_h single, gboolean is_input,
ml_tensors_info_h info = NULL;

ml_single_get_gst_info (single_h, is_input, &gst_info);
if (single_h->is_flexible) {
gst_info.format = _NNS_TENSOR_FORMAT_FLEXIBLE;
gst_info.num_tensors = 1U; /* TODO: Consider multiple input tensors filter */
}

_ml_tensors_info_create_from_gst (&info, &gst_info);

gst_tensors_info_free (&gst_info);
Expand Down Expand Up @@ -846,6 +853,7 @@ ml_single_create_handle (ml_nnfw_type_e nnfw)
single_h->output = NULL;
single_h->destroy_data_list = NULL;
single_h->invoking = FALSE;
single_h->is_flexible = FALSE;

gst_tensors_info_init (&single_h->in_info);
gst_tensors_info_init (&single_h->out_info);
Expand Down Expand Up @@ -1082,6 +1090,11 @@ ml_single_open_custom (ml_single_h * single, ml_single_preset * info)
status = ML_ERROR_STREAMS_PIPE;
goto error;
}
/* handle flexible single */
if (info->nnfw == ML_NNFW_TYPE_EXECUTORCH_LLAMA) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think if there are more flexible filter, we may revise this part later.
The llama2c filter I created doesn't seem to be scalable due to API limitations, so I won't add it now.

single_h->is_flexible = TRUE;
g_object_set (filter_obj, "invoke-dynamic", TRUE, NULL);
}

if (nnfw == ML_NNFW_TYPE_NNTR_INF) {
if (!in_tensors_info || !out_tensors_info) {
Expand Down Expand Up @@ -1318,6 +1331,11 @@ _ml_single_invoke_validate_data (ml_single_h single,
"The %d-th input tensor is not valid. There is no valid dimension metadata for this tensor.",
i);

if (single_h->is_flexible) {
/* Skip data size check for flexible */
continue;
}

raw_size = _model->tensors[i].size;
if (G_UNLIKELY (_data->tensors[i].size != raw_size))
_ml_error_report_return (ML_ERROR_INVALID_PARAMETER,
Expand Down Expand Up @@ -1957,6 +1975,7 @@ _ml_validate_model_file (const char *const *model,
case ML_NNFW_TYPE_ONNX_RUNTIME:
case ML_NNFW_TYPE_NCNN:
case ML_NNFW_TYPE_TENSORRT:
case ML_NNFW_TYPE_EXECUTORCH_LLAMA:
case ML_NNFW_TYPE_QNN:
/**
* We cannot check the file ext with NNFW.
Expand Down
46 changes: 46 additions & 0 deletions tests/capi/unittest_capi_inference_single.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3180,6 +3180,52 @@ TEST (nnstreamer_capi_singleshot, invoke_ncnn)
}
#endif /* ENABLE_NCNN */

/**
* @brief DISABLED Test to show executorch_llama filter usage
*/
TEST (nnstreamer_capi_singleshot, DISABLED_executorch_llama)
{
int status;
ml_single_h single;

status = ml_single_open (&single, "/path/to/pte,/path/to/tokienizer", NULL,
NULL, ML_NNFW_TYPE_EXECUTORCH_LLAMA, ML_NNFW_HW_ANY);
ASSERT_EQ (status, ML_ERROR_NONE);

/* prepare input data */
std::string prompt ("Once upon a time");
ml_tensors_info_h in_info;
ml_tensors_data_h in_data;
ml_tensor_dimension dim = { (unsigned int) prompt.size () + 1, 0 };

ml_tensors_info_create (&in_info);
ml_tensors_info_set_count (in_info, 1);
ml_tensors_info_set_tensor_type (in_info, 0, ML_TENSOR_TYPE_UINT8);
ml_tensors_info_set_tensor_dimension (in_info, 0, dim);

ml_tensors_data_create (in_info, &in_data);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

How about checking if in_data is created properly?

status = ml_tensors_data_create (in_info, &in_data);
EXPECT_EQ (status, ML_ERROR_NONE);

ml_tensors_data_set_tensor_data (in_data, 0, prompt.c_str (), prompt.size () + 1);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Let's check the result, how about you?
EXPECT_EQ (status, ML_ERROR_NONE);


/* invoke */
ml_tensors_data_h out_data;
status = ml_single_invoke (single, in_data, &out_data);
EXPECT_EQ (ML_ERROR_NONE, status);

char *result;
size_t result_size;
status = ml_tensors_data_get_tensor_data (out_data, 0U, (void **) &result, &result_size);
EXPECT_EQ (ML_ERROR_NONE, status);

g_info ("result: %s", result);
EXPECT_EQ (0, strncmp (result, prompt.c_str (), prompt.size ()));

/* free data */
ml_tensors_data_destroy (out_data);
ml_tensors_data_destroy (in_data);
ml_tensors_info_destroy (in_info);
ml_single_close (single);
}

/**
* @brief Test NNStreamer single shot (custom filter)
* @detail Run pipeline with custom filter with allocate in invoke, handle multi tensors.
Expand Down
Loading