5
from vllm.lora.request import LoRARequest
7
MODEL_PATH = "mistralai/Mixtral-8x7B-Instruct-v0.1"
10
def do_sample(llm, lora_path: str, lora_id: int):
12
"[system] Given a target sentence construct the underlying meaning representation\nof the input sentence as a single function with attributes and attribute\nvalues. This function should describe the target string accurately and the\nfunction must be one of the following ['inform', 'request', 'give_opinion',\n'confirm', 'verify_attribute', 'suggest', 'request_explanation',\n'recommend', 'request_attribute'].\n\nThe attributes must be one of the following:\n['name', 'exp_release_date', 'release_year', 'developer', 'esrb', 'rating',\n'genres', 'player_perspective', 'has_multiplayer', 'platforms',\n'available_on_steam', 'has_linux_release', 'has_mac_release', 'specifier'] [/system] [user] Here is the target sentence:\nSpellForce 3 is a pretty bad game. The developer Grimlore Games is clearly a bunch of no-talent hacks, and 2017 was a terrible year for games anyway. [/user] [assistant]",
13
"[system] Given a target sentence construct the underlying meaning representation\nof the input sentence as a single function with attributes and attribute\nvalues. This function should describe the target string accurately and the\nfunction must be one of the following ['inform', 'request', 'give_opinion',\n'confirm', 'verify_attribute', 'suggest', 'request_explanation',\n'recommend', 'request_attribute'].\n\nThe attributes must be one of the following:\n['name', 'exp_release_date', 'release_year', 'developer', 'esrb', 'rating',\n'genres', 'player_perspective', 'has_multiplayer', 'platforms',\n'available_on_steam', 'has_linux_release', 'has_mac_release', 'specifier'] [/system] [user] Here is the target sentence:\nI wanted to like Grimlore Games' 2017 entry, but in SpellForce 3 they just didn't get anything right. [/user] [assistant]",
14
"[system] Given a target sentence construct the underlying meaning representation\nof the input sentence as a single function with attributes and attribute\nvalues. This function should describe the target string accurately and the\nfunction must be one of the following ['inform', 'request', 'give_opinion',\n'confirm', 'verify_attribute', 'suggest', 'request_explanation',\n'recommend', 'request_attribute'].\n\nThe attributes must be one of the following:\n['name', 'exp_release_date', 'release_year', 'developer', 'esrb', 'rating',\n'genres', 'player_perspective', 'has_multiplayer', 'platforms',\n'available_on_steam', 'has_linux_release', 'has_mac_release', 'specifier'] [/system] [user] Here is the target sentence:\nBioShock is a good role-playing, action-adventure, shooter that released for PlayStation, Xbox, and PC in 2007. It is available on Steam, and it has a Mac release but not a Linux release. [/user] [assistant]",
16
sampling_params = vllm.SamplingParams(temperature=0, max_tokens=256)
17
outputs = llm.generate(
20
lora_request=LoRARequest(str(lora_id), lora_id, lora_path)
24
for output in outputs:
25
prompt = output.prompt
26
generated_text = output.outputs[0].text.strip()
27
generated_texts.append(generated_text)
28
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
29
return generated_texts
32
@pytest.mark.parametrize("tp_size", [4])
33
def test_mixtral_lora(mixtral_lora_files, tp_size):
34
if torch.cuda.device_count() < tp_size:
35
pytest.skip(f"Not enough GPUs for tensor parallelism {tp_size}")
37
llm = vllm.LLM(MODEL_PATH,
41
tensor_parallel_size=tp_size,
44
expected_lora_output = [
45
"give_opinion(name[SpellForce 3], release_year[2017], developer[Grimlore Games], rating[poor])",
46
"give_opinion(name[SpellForce 3], release_year[2017], developer[Grimlore Games], rating[poor])",
47
"inform(name[BioShock], release_year[2007], rating[good], genres[action-adventure, role-playing, shooter], platforms[PlayStation, Xbox, PC], available_on_steam[yes], has_linux_release[no], has_mac_release[yes])",
50
assert do_sample(llm, mixtral_lora_files,
51
lora_id=1) == expected_lora_output
52
assert do_sample(llm, mixtral_lora_files,
53
lora_id=2) == expected_lora_output