File size: 9,210 Bytes
b555842 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 | """
gui/pages/efficiency.py — ⚡ Efficiency Explorer
─────────────────────────────────────────────────────────────────────────────
energy_per_token, J/instruction, IPC correlation — find optimal configs.
─────────────────────────────────────────────────────────────────────────────
"""
import streamlit as st
import pandas as pd
import plotly.graph_objects as go
from gui.db import q, q1
from gui.config import PL, WF_COLORS
ACCENT = "#38bdf8"
def render(ctx: dict) -> None:
runs = ctx.get("runs", pd.DataFrame())
if runs.empty:
st.info("No run data available.")
return
df = runs.copy()
has_ept = "energy_per_token" in df.columns and df["energy_per_token"].notna().any()
has_epi = "energy_per_instruction" in df.columns
has_ipc = "ipc" in df.columns
has_task = "task_name" in df.columns
has_model = "model_name" in df.columns
# ── KPIs ──────────────────────────────────────────────────────────────────
df_e = df[df["energy_j"].notna() & (df["energy_j"] > 0)] if "energy_j" in df.columns else df
lin = df[df["workflow_type"] == "linear"]
age = df[df["workflow_type"] == "agentic"]
best_ept = df[df["energy_per_token"] > 0]["energy_per_token"].min() if has_ept else 0
avg_ept_l = lin[lin["energy_per_token"] > 0]["energy_per_token"].mean() if has_ept and not lin.empty else 0
avg_ept_a = age[age["energy_per_token"] > 0]["energy_per_token"].mean() if has_ept and not age.empty else 0
avg_ipc = df["ipc"].mean() if has_ipc else 0
st.markdown(
f"<div style='padding:14px 20px;"
f"background:linear-gradient(135deg,{ACCENT}14,{ACCENT}06);"
f"border:1px solid {ACCENT}33;border-radius:12px;margin-bottom:20px;'>"
f"<div style='font-size:11px;font-weight:700;color:{ACCENT};"
f"text-transform:uppercase;letter-spacing:.1em;margin-bottom:10px;'>"
f"Efficiency Explorer — {len(df)} runs</div>"
f"<div style='display:grid;grid-template-columns:repeat(4,1fr);gap:12px;'>"
+ "".join([
f"<div><div style='font-size:16px;font-weight:700;color:{c};"
f"font-family:IBM Plex Mono,monospace;line-height:1;'>{v}</div>"
f"<div style='font-size:9px;color:#94a3b8;text-transform:uppercase;"
f"letter-spacing:.08em;margin-top:3px;'>{l}</div></div>"
for v, l, c in [
(f"{best_ept:.5f}J", "Best J/token", "#22c55e"),
(f"{avg_ept_l:.5f}J", "Avg J/token linear", WF_COLORS["linear"]),
(f"{avg_ept_a:.5f}J", "Avg J/token agentic", WF_COLORS["agentic"]),
(f"{avg_ipc:.2f}", "Avg IPC", ACCENT),
]
])
+ "</div></div>",
unsafe_allow_html=True)
# ── Energy per token by model × workflow ──────────────────────────────────
if has_ept and has_model:
st.markdown(
f"<div style='font-size:11px;font-weight:600;color:{ACCENT};"
f"text-transform:uppercase;letter-spacing:.1em;margin-bottom:8px;'>"
f"Energy per token — model × workflow</div>",
unsafe_allow_html=True)
ept_df = df[df["energy_per_token"].notna() & (df["energy_per_token"] > 0)]
model_ept = ept_df.groupby(
["model_name","workflow_type"])["energy_per_token"].mean().reset_index()
fig = go.Figure()
for wf, clr in WF_COLORS.items():
sub = model_ept[model_ept["workflow_type"] == wf]
if sub.empty: continue
fig.add_trace(go.Bar(
x=sub["model_name"], y=sub["energy_per_token"],
name=wf, marker_color=clr, marker_line_width=0))
fig.update_layout(**PL, height=260, barmode="group",
xaxis_title="Model",
yaxis_title="J / token",
xaxis_tickangle=-30)
st.plotly_chart(fig, use_container_width=True, key="eff_ept_model")
# ── Efficiency scatter: IPC vs energy ─────────────────────────────────────
col1, col2 = st.columns(2)
with col1:
st.markdown(
f"<div style='font-size:11px;font-weight:600;color:{ACCENT};"
f"text-transform:uppercase;letter-spacing:.1em;margin-bottom:8px;'>"
f"IPC vs Energy — efficiency frontier</div>",
unsafe_allow_html=True)
fig2 = go.Figure()
if has_ipc and "energy_j" in df.columns:
for wf, clr in WF_COLORS.items():
sub = df[df["workflow_type"] == wf]
sub = sub[sub["ipc"].notna() & sub["energy_j"].notna() &
(sub["ipc"] > 0) & (sub["energy_j"] > 0)]
if sub.empty: continue
fig2.add_trace(go.Scatter(
x=sub["ipc"], y=sub["energy_j"],
mode="markers", name=wf,
marker=dict(color=clr, size=5, opacity=0.6)))
fig2.update_layout(**PL, height=260,
xaxis_title="IPC (higher = better)",
yaxis_title="Energy (J) (lower = better)")
st.plotly_chart(fig2, use_container_width=True, key="eff_ipc_energy")
with col2:
st.markdown(
f"<div style='font-size:11px;font-weight:600;color:{ACCENT};"
f"text-transform:uppercase;letter-spacing:.1em;margin-bottom:8px;'>"
f"Energy per token vs total tokens</div>",
unsafe_allow_html=True)
fig3 = go.Figure()
if has_ept and "total_tokens" in df.columns:
ept_sub = df[df["energy_per_token"].notna() &
df["total_tokens"].notna() &
(df["energy_per_token"] > 0) &
(df["total_tokens"] > 0)]
for wf, clr in WF_COLORS.items():
sub = ept_sub[ept_sub["workflow_type"] == wf]
if sub.empty: continue
fig3.add_trace(go.Scatter(
x=sub["total_tokens"], y=sub["energy_per_token"],
mode="markers", name=wf,
marker=dict(color=clr, size=5, opacity=0.6)))
fig3.update_layout(**PL, height=260,
xaxis_title="Total tokens",
yaxis_title="J / token")
st.plotly_chart(fig3, use_container_width=True, key="eff_ept_tokens")
# ── Efficiency by task ────────────────────────────────────────────────────
if has_task and has_ept:
st.markdown(
f"<div style='font-size:11px;font-weight:600;color:{ACCENT};"
f"text-transform:uppercase;letter-spacing:.1em;margin:16px 0 8px;'>"
f"Efficiency by task — J/token ranking</div>",
unsafe_allow_html=True)
task_eff = df[df["energy_per_token"] > 0].groupby("task_name").agg(
avg_ept=("energy_per_token", "mean"),
avg_ipc=("ipc", "mean"),
runs =("energy_j", "count"),
).reset_index().sort_values("avg_ept")
fig4 = go.Figure(go.Bar(
x=task_eff["avg_ept"],
y=task_eff["task_name"],
orientation="h",
marker_color=ACCENT, marker_line_width=0,
text=[f"{v:.5f}J" for v in task_eff["avg_ept"]],
textposition="outside", textfont=dict(size=9)))
fig4.update_layout(
**{**PL, "margin": dict(l=160, r=80, t=10, b=30)},
height=max(200, len(task_eff) * 28),
xaxis_title="Avg J / token",
showlegend=False)
st.plotly_chart(fig4, use_container_width=True, key="eff_task_rank")
# ── Optimal config finder ─────────────────────────────────────────────────
st.markdown(
f"<div style='font-size:11px;font-weight:600;color:{ACCENT};"
f"text-transform:uppercase;letter-spacing:.1em;margin:16px 0 8px;'>"
f"Top 10 most efficient runs</div>",
unsafe_allow_html=True)
if has_ept:
top10 = df[df["energy_per_token"] > 0].nsmallest(10, "energy_per_token")[[
"run_id" if "run_id" in df.columns else df.columns[0],
"workflow_type", "model_name", "task_name",
"energy_j", "energy_per_token", "ipc",
]].round(6) if "energy_j" in df.columns else pd.DataFrame()
if not top10.empty:
st.dataframe(top10, use_container_width=True)
|