A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 17, 09:15 UTC
Change plotting script
Woosuk Kwon committed
Apr 17, 2023 at 04:49 UTC
aa50b17ca776f8c69a793787d0ce06dfa4671884
1 file changed
+11
-2
plot/plot_normalized_latency.py
+11
-2
@@ -16,9 +16,9 @@ SYSTEMS = [
16
17
SYSTEM_TO_LABEL = {
18
'orca-constant': 'Orca (Max)',
19
- 'orca-power2': 'Orca (Next power of 2)',
19
+ 'orca-power2': 'Orca (Pow2)',
20
'orca-oracle': 'Orca (Oracle)',
21
- 'cacheflow': 'CacheFlow',
21
+ 'cacheflow': 'KVFlow',
22
}
23
24
SYSTEM_TO_COLOR = {
@@ -178,6 +178,15 @@ def plot_normalized_latency(
178
else:
179
plt.ylim(bottom=0, top=ylim)
180
181
+ handles, labels = plt.gca().get_legend_handles_labels()
182
+ handles = reversed(handles)
183
+ labels = reversed(labels)
184
+
185
+ plt.legend(
186
+ handles, labels,
187
+ ncol=4, fontsize=12, loc='upper center', bbox_to_anchor=(0.5, 1.15),
188
+ columnspacing=0.5, handletextpad=0.5, handlelength=1.5, frameon=False, borderpad=0)
189
+
190
# Save figure.
191
model, tp = get_model(exp_dir)
192
sampling = get_sampling(exp_dir)