From 5fcf5124ca112aa0a5f17ab66bd1e8f06522739a Mon Sep 17 00:00:00 2001
From: dyastremsky <58150256+dyastremsky@users.noreply.github.com>
Date: Wed, 6 Mar 2024 17:12:42 -0800
Subject: [PATCH] Add formatting for GenAi-PA report (#491)

---
 .../genai-pa/genai_pa/llm_metrics.py          | 35 +++++++++++++++++++
 .../perf_analyzer/genai-pa/genai_pa/main.py   |  4 +--
 2 files changed, 37 insertions(+), 2 deletions(-)

diff --git a/src/c++/perf_analyzer/genai-pa/genai_pa/llm_metrics.py b/src/c++/perf_analyzer/genai-pa/genai_pa/llm_metrics.py
index 80230d176..29badd843 100755
--- a/src/c++/perf_analyzer/genai-pa/genai_pa/llm_metrics.py
+++ b/src/c++/perf_analyzer/genai-pa/genai_pa/llm_metrics.py
@@ -33,6 +33,8 @@
 
 import numpy as np
 from genai_pa.utils import load_json
+from rich.console import Console
+from rich.table import Table
 
 # Silence tokenizer warning on import
 with contextlib.redirect_stdout(io.StringIO()) as stdout, contextlib.redirect_stderr(
@@ -121,6 +123,38 @@ def __repr__(self):
         attr_strs = ",".join([f"{k}={v}" for k, v in self.__dict__.items()])
         return f"Statistics({attr_strs})"
 
+    def _is_time_field(self, field: str):
+        time_fields = [
+            "inter_token_latency",
+            "time_to_first_token",
+            "end_to_end_latency",
+        ]
+        return field in time_fields
+
+    def pretty_print(self):
+        table = Table(title="PA LLM Metrics")
+
+        table.add_column("Statistic", justify="right", style="cyan", no_wrap=True)
+        stats = ["avg", "min", "max", "p99", "p95", "p90", "p75", "p50", "p25"]
+        for stat in stats:
+            table.add_column(stat, justify="right", style="green")
+
+        metrics = ["inter_token_latency", "time_to_first_token"]
+        for metric in metrics:
+            formatted_metric = metric.replace("_", " ").capitalize()
+            is_time_field = self._is_time_field(metric)
+            if is_time_field:
+                formatted_metric += " (ns)"
+            row_values = [formatted_metric]
+
+            for stat in stats:
+                value = self.__dict__.get(f"{stat}_{metric}", -1)
+                row_values.append("{:,.0f}".format(value))
+            table.add_row(*row_values)
+
+        console = Console()
+        console.print(table)
+
 
 class LLMProfileData:
     """A class that calculates and aggregates all the LLM performance statistics
@@ -141,6 +175,7 @@ class LLMProfileData:
       >>> stats = pd.get_statistics(infer_mode="concurrency", level=10)
       >>>
       >>> print(stats)  # output: Statistics(avg_time_to_first_token=...)
+      >>> stats.pretty_print()  # Output: time_to_first_token_s: ...
     """
 
     def __init__(self, filename: str, tokenizer: AutoTokenizer) -> None:
diff --git a/src/c++/perf_analyzer/genai-pa/genai_pa/main.py b/src/c++/perf_analyzer/genai-pa/genai_pa/main.py
index 3a9246c55..df4ca1bc0 100755
--- a/src/c++/perf_analyzer/genai-pa/genai_pa/main.py
+++ b/src/c++/perf_analyzer/genai-pa/genai_pa/main.py
@@ -76,8 +76,8 @@ def report_output(metrics: LLMProfileData, args):
         raise GenAiPAException(
             "Neither concurrency_range nor request_rate_range was found in args when reporting metrics"
         )
-    # TODO: metrics reporter class that consumes Stats class for nicer formatting
-    print(metrics.get_statistics(infer_mode, int(load_level)))
+    stats = metrics.get_statistics(infer_mode, int(load_level))
+    stats.pretty_print()
 
 
 # Separate function that can raise exceptions used for testing