triton-model-analyzer 1.48.0__py3-none-any.whl

model_analyzer/record/types/perf_client_response_wait.py

@@ -0,0 +1,97 @@
+#!/usr/bin/env python3
+
+# Copyright 2020-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from functools import total_ordering
+
+from model_analyzer.record.record import DecreasingRecord
+
+
+@total_ordering
+class PerfClientResponseWait(DecreasingRecord):
+    """
+    A record for perf_analyzer
+    metric 'Client response wait'
+    """
+
+    tag = "perf_client_response_wait"
+
+    def __init__(self, value, timestamp=0):
+        """
+        Parameters
+        ----------
+        value : float
+            the latency extracted from the perf analyzer output
+        timestamp : float
+            Elapsed time from start of program
+        """
+
+        super().__init__(value, timestamp)
+
+    @staticmethod
+    def header(aggregation_tag=False):
+        """
+        Parameters
+        ----------
+        aggregation_tag: bool
+            An optional tag that may be displayed
+            as part of the header indicating that
+            this record has been aggregated using
+            max, min or average etc.
+
+        Returns
+        -------
+        str
+            The full name of the
+            metric.
+        """
+
+        return "Response Wait Time (ms)"
+
+    def __eq__(self, other):
+        """
+        Allows checking for
+        equality between two records
+        """
+
+        return self.value() == other.value()
+
+    def __lt__(self, other):
+        """
+        Allows checking if
+        this record is less than
+        the other
+        """
+
+        return self.value() > other.value()
+
+    def __add__(self, other):
+        """
+        Allows adding two records together
+        to produce a brand new record.
+        """
+
+        return PerfClientResponseWait(value=(self.value() + other.value()))
+
+    def __sub__(self, other):
+        """
+        Allows subbing two records together
+        to produce a brand new record.
+
+        ** Note this does reverse subtraction because
+            of the inverted nature of latency (lower is better)
+        """
+
+        return PerfClientResponseWait(value=(other.value() - self.value()))

model_analyzer/record/types/perf_client_send_recv.py

@@ -0,0 +1,97 @@
+#!/usr/bin/env python3
+
+# Copyright 2020-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from functools import total_ordering
+
+from model_analyzer.record.record import DecreasingRecord
+
+
+@total_ordering
+class PerfClientSendRecv(DecreasingRecord):
+    """
+    A record for perf_analyzer
+    metric 'Client send/recv'
+    """
+
+    tag = "perf_client_send_recv"
+
+    def __init__(self, value, timestamp=0):
+        """
+        Parameters
+        ----------
+        value : float
+            the latency extracted from the perf analyzer output
+        timestamp : float
+            Elapsed time from start of program
+        """
+
+        super().__init__(value, timestamp)
+
+    @staticmethod
+    def header(aggregation_tag=False):
+        """
+        Parameters
+        ----------
+        aggregation_tag: bool
+            An optional tag that may be displayed
+            as part of the header indicating that
+            this record has been aggregated using
+            max, min or average etc.
+
+        Returns
+        -------
+        str
+            The full name of the
+            metric.
+        """
+
+        return "Client Send/Recv (ms)"
+
+    def __eq__(self, other):
+        """
+        Allows checking for
+        equality between two records
+        """
+
+        return self.value() == other.value()
+
+    def __lt__(self, other):
+        """
+        Allows checking if
+        this record is less than
+        the other
+        """
+
+        return self.value() > other.value()
+
+    def __add__(self, other):
+        """
+        Allows adding two records together
+        to produce a brand new record.
+        """
+
+        return PerfClientSendRecv(value=(self.value() + other.value()))
+
+    def __sub__(self, other):
+        """
+        Allows subbing two records together
+        to produce a brand new record.
+
+        ** Note this does reverse subtraction because
+            of the inverted nature of latency (lower is better)
+        """
+
+        return PerfClientSendRecv(value=(other.value() - self.value()))

model_analyzer/record/types/perf_latency.py

@@ -0,0 +1,111 @@
+#!/usr/bin/env python3
+
+# Copyright 2021-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+from functools import total_ordering
+
+from model_analyzer.record.record import DecreasingRecord
+
+
+@total_ordering
+class PerfLatency(DecreasingRecord):
+    """
+    A record for perf_analyzer latency metric
+    """
+
+    tag = "perf_latency"
+    _printed_deprecate_warning = False
+
+    def __init__(self, value, timestamp=0):
+        """
+        Parameters
+        ----------
+        value : float
+            the latency extracted from the perf analyzer output
+        timestamp : float
+            Elapsed time from start of program
+        """
+
+        super().__init__(value, timestamp)
+
+    @classmethod
+    def header(cls, aggregation_tag=False):
+        """
+        Parameters
+        ----------
+        aggregation_tag: bool
+            An optional tag that may be displayed
+            as part of the header indicating that
+            this record has been aggregated using
+            max, min or average etc.
+
+        Returns
+        -------
+        str
+            The full name of the
+            metric.
+        """
+
+        cls._print_deprecate_warning()
+        return "p99 Latency (ms)"
+
+    def __eq__(self, other):
+        """
+        Allows checking for
+        equality between two records
+        """
+
+        self._print_deprecate_warning()
+        return self.value() == other.value()
+
+    def __lt__(self, other):
+        """
+        Allows checking if
+        this record is less than
+        the other
+        """
+
+        self._print_deprecate_warning()
+        return self.value() > other.value()
+
+    def __add__(self, other):
+        """
+        Allows adding two records together
+        to produce a brand new record.
+        """
+
+        self._print_deprecate_warning()
+        return self.__class__(value=(self.value() + other.value()))
+
+    def __sub__(self, other):
+        """
+        Allows subbing two records together
+        to produce a brand new record.
+
+        ** Note this does reverse subtraction because
+            of the inverted nature of latency (lower is better)
+        """
+
+        self._print_deprecate_warning()
+        return self.__class__(value=(other.value() - self.value()))
+
+    @classmethod
+    def _print_deprecate_warning(cls):
+        if not cls._printed_deprecate_warning:
+            logging.warning(
+                "`perf_latency` is deprecated, use `perf_latency_p99` instead."
+            )
+            cls._printed_deprecate_warning = True

model_analyzer/record/types/perf_latency_avg.py

@@ -0,0 +1,60 @@
+#!/usr/bin/env python3
+
+# Copyright 2021-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from functools import total_ordering
+
+from model_analyzer.record.types.perf_latency_base import PerfLatencyBase
+
+
+@total_ordering
+class PerfLatencyAvg(PerfLatencyBase):
+    """
+    A record for perf_analyzer latency metric
+    """
+
+    tag = "perf_latency_avg"
+
+    def __init__(self, value, timestamp=0):
+        """
+        Parameters
+        ----------
+        value : float
+            the latency extracted from the perf analyzer output
+        timestamp : float
+            Elapsed time from start of program
+        """
+
+        super().__init__(value, timestamp)
+
+    @classmethod
+    def header(cls, aggregation_tag=False):
+        """
+        Parameters
+        ----------
+        aggregation_tag: bool
+            An optional tag that may be displayed
+            as part of the header indicating that
+            this record has been aggregated using
+            max, min or average etc.
+
+        Returns
+        -------
+        str
+            The full name of the
+            metric.
+        """
+
+        return "Avg Latency (ms)"

model_analyzer/record/types/perf_latency_base.py

@@ -0,0 +1,74 @@
+#!/usr/bin/env python3
+
+# Copyright 2021-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from functools import total_ordering
+
+from model_analyzer.record.record import DecreasingRecord
+
+
+@total_ordering
+class PerfLatencyBase(DecreasingRecord):
+    """
+    A base class for perf_analyzer latency metric
+    """
+
+    def __init__(self, value, timestamp=0):
+        """
+        Parameters
+        ----------
+        value : float
+            the latency extracted from the perf analyzer output
+        timestamp : float
+            Elapsed time from start of program
+        """
+
+        super().__init__(value, timestamp)
+
+    def __eq__(self, other):
+        """
+        Allows checking for
+        equality between two records
+        """
+
+        return self.value() == other.value()
+
+    def __lt__(self, other):
+        """
+        Allows checking if
+        this record is less than
+        the other
+        """
+
+        return self.value() > other.value()
+
+    def __add__(self, other):
+        """
+        Allows adding two records together
+        to produce a brand new record.
+        """
+
+        return self.__class__(value=(self.value() + other.value()))
+
+    def __sub__(self, other):
+        """
+        Allows subbing two records together
+        to produce a brand new record.
+
+        ** Note this does reverse subtraction because
+            of the inverted nature of latency (lower is better)
+        """
+
+        return self.__class__(value=(other.value() - self.value()))

model_analyzer/record/types/perf_latency_p90.py

@@ -0,0 +1,60 @@
+#!/usr/bin/env python3
+
+# Copyright 2021-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from functools import total_ordering
+
+from model_analyzer.record.types.perf_latency_base import PerfLatencyBase
+
+
+@total_ordering
+class PerfLatencyP90(PerfLatencyBase):
+    """
+    A record for perf_analyzer latency metric
+    """
+
+    tag = "perf_latency_p90"
+
+    def __init__(self, value, timestamp=0):
+        """
+        Parameters
+        ----------
+        value : float
+            the latency extracted from the perf analyzer output
+        timestamp : float
+            Elapsed time from start of program
+        """
+
+        super().__init__(value, timestamp)
+
+    @classmethod
+    def header(cls, aggregation_tag=False):
+        """
+        Parameters
+        ----------
+        aggregation_tag: bool
+            An optional tag that may be displayed
+            as part of the header indicating that
+            this record has been aggregated using
+            max, min or average etc.
+
+        Returns
+        -------
+        str
+            The full name of the
+            metric.
+        """
+
+        return "p90 Latency (ms)"

model_analyzer/record/types/perf_latency_p95.py

@@ -0,0 +1,60 @@
+#!/usr/bin/env python3
+
+# Copyright 2021-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from functools import total_ordering
+
+from model_analyzer.record.types.perf_latency_base import PerfLatencyBase
+
+
+@total_ordering
+class PerfLatencyP95(PerfLatencyBase):
+    """
+    A record for perf_analyzer latency metric
+    """
+
+    tag = "perf_latency_p95"
+
+    def __init__(self, value, timestamp=0):
+        """
+        Parameters
+        ----------
+        value : float
+            the latency extracted from the perf analyzer output
+        timestamp : float
+            Elapsed time from start of program
+        """
+
+        super().__init__(value, timestamp)
+
+    @classmethod
+    def header(cls, aggregation_tag=False):
+        """
+        Parameters
+        ----------
+        aggregation_tag: bool
+            An optional tag that may be displayed
+            as part of the header indicating that
+            this record has been aggregated using
+            max, min or average etc.
+
+        Returns
+        -------
+        str
+            The full name of the
+            metric.
+        """
+
+        return "p95 Latency (ms)"

model_analyzer/record/types/perf_latency_p99.py

@@ -0,0 +1,60 @@
+#!/usr/bin/env python3
+
+# Copyright 2021-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from functools import total_ordering
+
+from model_analyzer.record.types.perf_latency_base import PerfLatencyBase
+
+
+@total_ordering
+class PerfLatencyP99(PerfLatencyBase):
+    """
+    A record for perf_analyzer latency metric
+    """
+
+    tag = "perf_latency_p99"
+
+    def __init__(self, value, timestamp=0):
+        """
+        Parameters
+        ----------
+        value : float
+            the latency extracted from the perf analyzer output
+        timestamp : float
+            Elapsed time from start of program
+        """
+
+        super().__init__(value, timestamp)
+
+    @classmethod
+    def header(cls, aggregation_tag=False):
+        """
+        Parameters
+        ----------
+        aggregation_tag: bool
+            An optional tag that may be displayed
+            as part of the header indicating that
+            this record has been aggregated using
+            max, min or average etc.
+
+        Returns
+        -------
+        str
+            The full name of the
+            metric.
+        """
+
+        return "p99 Latency (ms)"