Spaces:
Running
Running
Commit
·
9feeb78
1
Parent(s):
e70447d
use reduction for latency instead of speedup
Browse files
run.py
CHANGED
@@ -149,10 +149,10 @@ def postprocess_table(table, experiment_name):
|
|
149 |
table = table.set_index("experiment_name")
|
150 |
table.reset_index(inplace=True)
|
151 |
if "forward.latency(s)" in table.columns:
|
152 |
-
table["forward.latency.
|
153 |
-
|
154 |
) * 100
|
155 |
-
table["forward.latency.
|
156 |
|
157 |
if "forward.throughput(samples/s)" in table.columns:
|
158 |
table["forward.throughput.speedup(%)"] = (
|
@@ -161,16 +161,16 @@ def postprocess_table(table, experiment_name):
|
|
161 |
table["forward.throughput.speedup(%)"] = table["forward.throughput.speedup(%)"].round(2)
|
162 |
|
163 |
if "forward.peak_memory(MB)" in table.columns:
|
164 |
-
table["forward.peak_memory.
|
165 |
-
|
166 |
) * 100
|
167 |
-
table["forward.peak_memory.
|
168 |
|
169 |
if "generate.latency(s)" in table.columns:
|
170 |
-
table["generate.latency.
|
171 |
-
|
172 |
) * 100
|
173 |
-
table["generate.latency.
|
174 |
|
175 |
if "generate.throughput(tokens/s)" in table.columns:
|
176 |
table["generate.throughput.speedup(%)"] = (
|
@@ -179,9 +179,9 @@ def postprocess_table(table, experiment_name):
|
|
179 |
table["generate.throughput.speedup(%)"] = table["generate.throughput.speedup(%)"].round(2)
|
180 |
|
181 |
if "generate.peak_memory(MB)" in table.columns:
|
182 |
-
table["generate.peak_memory.
|
183 |
-
|
184 |
) * 100
|
185 |
-
table["generate.peak_memory.
|
186 |
|
187 |
return table
|
|
|
149 |
table = table.set_index("experiment_name")
|
150 |
table.reset_index(inplace=True)
|
151 |
if "forward.latency(s)" in table.columns:
|
152 |
+
table["forward.latency.reduction(%)"] = (
|
153 |
+
table["forward.latency(s)"] / table["forward.latency(s)"].iloc[0] - 1
|
154 |
) * 100
|
155 |
+
table["forward.latency.reduction(%)"] = table["forward.latency.reduction(%)"].round(2)
|
156 |
|
157 |
if "forward.throughput(samples/s)" in table.columns:
|
158 |
table["forward.throughput.speedup(%)"] = (
|
|
|
161 |
table["forward.throughput.speedup(%)"] = table["forward.throughput.speedup(%)"].round(2)
|
162 |
|
163 |
if "forward.peak_memory(MB)" in table.columns:
|
164 |
+
table["forward.peak_memory.reduction(%)"] = (
|
165 |
+
table["forward.peak_memory(MB)"] / table["forward.peak_memory(MB)"].iloc[0] - 1
|
166 |
) * 100
|
167 |
+
table["forward.peak_memory.reduction(%)"] = table["forward.peak_memory.savings(%)"].round(2)
|
168 |
|
169 |
if "generate.latency(s)" in table.columns:
|
170 |
+
table["generate.latency.reduction(%)"] = (
|
171 |
+
table["generate.latency(s)"] / table["generate.latency(s)"].iloc[0] - 1
|
172 |
) * 100
|
173 |
+
table["generate.latency.reduction(%)"] = table["generate.latency.reduction(%)"].round(2)
|
174 |
|
175 |
if "generate.throughput(tokens/s)" in table.columns:
|
176 |
table["generate.throughput.speedup(%)"] = (
|
|
|
179 |
table["generate.throughput.speedup(%)"] = table["generate.throughput.speedup(%)"].round(2)
|
180 |
|
181 |
if "generate.peak_memory(MB)" in table.columns:
|
182 |
+
table["generate.peak_memory.reduction(%)"] = (
|
183 |
+
table["generate.peak_memory(MB)"] / table["generate.peak_memory(MB)"].iloc[0] - 1
|
184 |
) * 100
|
185 |
+
table["generate.peak_memory.reduction(%)"] = table["generate.peak_memory.reduction(%)"].round(2)
|
186 |
|
187 |
return table
|