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

model_analyzer/record/types/inter_token_latency_base.py

@@ -0,0 +1,74 @@
+#!/usr/bin/env python3
+
+# Copyright 2024, 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 InterTokenLatencyBase(DecreasingRecord):
+    """
+    A record for perf_analyzer Inter token 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/inter_token_latency_max.py

@@ -0,0 +1,60 @@
+#!/usr/bin/env python3
+
+# Copyright 2024, 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.inter_token_latency_base import InterTokenLatencyBase
+
+
+@total_ordering
+class InterTokenLatencyMax(InterTokenLatencyBase):
+    """
+    A record for perf_analyzer Inter token latency metric
+    """
+
+    tag = "inter_token_latency_max"
+
+    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 "Max Inter Token Latency (ms)"

model_analyzer/record/types/inter_token_latency_min.py

@@ -0,0 +1,60 @@
+#!/usr/bin/env python3
+
+# Copyright 2024, 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.inter_token_latency_base import InterTokenLatencyBase
+
+
+@total_ordering
+class InterTokenLatencyMin(InterTokenLatencyBase):
+    """
+    A record for perf_analyzer Inter token latency metric
+    """
+
+    tag = "inter_token_latency_min"
+
+    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 "Min Inter Token Latency (ms)"

model_analyzer/record/types/inter_token_latency_p25.py

@@ -0,0 +1,60 @@
+#!/usr/bin/env python3
+
+# Copyright 2024, 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.inter_token_latency_base import InterTokenLatencyBase
+
+
+@total_ordering
+class InterTokenLatencyP25(InterTokenLatencyBase):
+    """
+    A record for perf_analyzer Inter token latency metric
+    """
+
+    tag = "inter_token_latency_p25"
+
+    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 "p25 Inter Token Latency (ms)"

model_analyzer/record/types/inter_token_latency_p50.py

@@ -0,0 +1,60 @@
+#!/usr/bin/env python3
+
+# Copyright 2024, 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.inter_token_latency_base import InterTokenLatencyBase
+
+
+@total_ordering
+class InterTokenLatencyP50(InterTokenLatencyBase):
+    """
+    A record for perf_analyzer Inter token latency metric
+    """
+
+    tag = "inter_token_latency_p50"
+
+    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 "p50 Inter Token Latency (ms)"

model_analyzer/record/types/inter_token_latency_p75.py

@@ -0,0 +1,60 @@
+#!/usr/bin/env python3
+
+# Copyright 2024, 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.inter_token_latency_base import InterTokenLatencyBase
+
+
+@total_ordering
+class InterTokenLatencyP75(InterTokenLatencyBase):
+    """
+    A record for perf_analyzer Inter token latency metric
+    """
+
+    tag = "inter_token_latency_p75"
+
+    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 "p75 Inter Token Latency (ms)"

model_analyzer/record/types/inter_token_latency_p90.py

@@ -0,0 +1,60 @@
+#!/usr/bin/env python3
+
+# Copyright 2024, 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.inter_token_latency_base import InterTokenLatencyBase
+
+
+@total_ordering
+class InterTokenLatencyP90(InterTokenLatencyBase):
+    """
+    A record for perf_analyzer Inter token latency metric
+    """
+
+    tag = "inter_token_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 Inter Token Latency (ms)"

model_analyzer/record/types/inter_token_latency_p95.py

@@ -0,0 +1,60 @@
+#!/usr/bin/env python3
+
+# Copyright 2024, 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.inter_token_latency_base import InterTokenLatencyBase
+
+
+@total_ordering
+class InterTokenLatencyP95(InterTokenLatencyBase):
+    """
+    A record for perf_analyzer Inter token latency metric
+    """
+
+    tag = "inter_token_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 Inter Token Latency (ms)"

model_analyzer/record/types/inter_token_latency_p99.py

@@ -0,0 +1,60 @@
+#!/usr/bin/env python3
+
+# Copyright 2024, 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.inter_token_latency_base import InterTokenLatencyBase
+
+
+@total_ordering
+class InterTokenLatencyP99(InterTokenLatencyBase):
+    """
+    A record for perf_analyzer Inter token latency metric
+    """
+
+    tag = "inter_token_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 Inter Token Latency (ms)"

model_analyzer/record/types/output_token_throughput.py

@@ -0,0 +1,105 @@
+#!/usr/bin/env python3
+
+# Copyright 2024, 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 IncreasingRecord
+
+
+@total_ordering
+class OutputTokenThroughput(IncreasingRecord):
+    """
+    A record for perf_analyzer
+    metric 'Output Token Throughput'
+    """
+
+    tag = "output_token_throughput"
+
+    def __init__(self, value, timestamp=0):
+        """
+        Parameters
+        ----------
+        value : float
+            The throughput from the perf analyzer output
+        timestamp : float
+            Elapsed time from start of program
+        """
+
+        super().__init__(value, timestamp)
+
+    @staticmethod
+    def value_function():
+        """
+        Returns the total value from a list
+
+        Returns
+        -------
+        Total value of the list
+        """
+        return sum
+
+    @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 "Output Token Throughput (infer/sec)"
+
+    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 subtracting two records together
+        to produce a brand new record.
+        """
+
+        return self.__class__(value=(self.value() - other.value()))