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 16, 17:00 UTC
[CI] Change test input in Gemma LoRA test (#8163)
Woosuk Kwon committed
Sep 4, 2024 at 13:05 UTC
561d6f8077c54c7af5dbf2ed92131ce9f7d9b56b
1 file changed
+3
-2
tests/lora/test_gemma.py
+3
-2
@@ -13,7 +13,7 @@ def do_sample(llm: vllm.LLM, lora_path: str, lora_id: int) -> List[str]:
13
prompts = [
14
"Quote: Imagination is",
15
"Quote: Be yourself;",
16
- "Quote: So many books,",
16
+ "Quote: Painting is poetry that is seen rather than felt,",
17
]
18
sampling_params = vllm.SamplingParams(temperature=0, max_tokens=32)
19
outputs = llm.generate(
@@ -41,7 +41,8 @@ def test_gemma_lora(gemma_lora_files):
41
expected_lora_output = [
42
"more important than knowledge.\nAuthor: Albert Einstein\n",
43
"everyone else is already taken.\nAuthor: Oscar Wilde\n",
44
- "so little time\nAuthor: Frank Zappa\n",
44
+ "and poetry is painting that is felt rather than seen.\n"
45
+ "Author: Leonardo da Vinci\n",
46
]
47
48
output1 = do_sample(llm, gemma_lora_files, lora_id=1)