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, 00:30 UTC
[Bugfix] Add warmup for prefix caching example (#5235)
Zhuohan Li committed
Jun 3, 2024 at 19:36 UTC
bd0e7802e09e40060b857b85227fd93a6a739467
1 file changed
+4
-2
examples/offline_inference_with_prefix.py
+4
-2
@@ -51,8 +51,10 @@ for output in outputs:
51
52
print("-" * 80)
53
54
-# The llm.generate call will batch all prompts and send the batch at once
55
-# if resources allow.
54
+# Warmup so that the shared prompt's KV cache is computed.
55
+prefix_cached_llm.generate(generating_prompts[0], sampling_params)
56
+
57
+# Generate with prefix caching.
58
start_time_cached = time()
59
outputs = prefix_cached_llm.generate(generating_prompts, sampling_params)
60
duration_cached = time() - start_time_cached