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, 10:45 UTC
[Model] add a bunch of supported lora modules for mixtral (#9008)
Signed-off-by: Prashant Gupta <prashantgupta@us.ibm.com>
Prashant Gupta committed
Oct 4, 2024 at 09:24 UTC
9ade8bbc8dc63c03b9399f05e85a0d0ddc6f5788
3 files changed
+69
-20
tests/lora/conftest.py
+5
@@ -173,6 +173,11 @@ def mixtral_lora_files():
173
return snapshot_download(repo_id="SangBinCho/mixtral-lora")
174
175
176
+@pytest.fixture(scope="session")
177
+def mixtral_lora_files_all_target_modules():
178
+ return snapshot_download(repo_id="dyang415/mixtral-lora-v0")
179
+
180
+
181
@pytest.fixture(scope="session")
182
def gemma_lora_files():
183
return snapshot_download(repo_id="wskwon/gemma-7b-test-lora")
tests/lora/test_mixtral.py
+62
-16
@@ -9,12 +9,9 @@ from vllm.lora.request import LoRARequest
9
MODEL_PATH = "mistralai/Mixtral-8x7B-Instruct-v0.1"
10
11
12
-def do_sample(llm: vllm.LLM, lora_path: str, lora_id: int) -> List[str]:
13
- prompts = [
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:\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]", # noqa: E501
15
- "[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]", # noqa: E501
16
- "[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]", # noqa: E501
17
- ]
12
+def do_sample(llm: vllm.LLM, lora_path: str, lora_id: int,
13
+ prompts: List[str]) -> List[str]:
14
+
15
sampling_params = vllm.SamplingParams(temperature=0, max_tokens=256)
16
outputs = llm.generate(
17
prompts,
@@ -33,22 +30,71 @@ def do_sample(llm: vllm.LLM, lora_path: str, lora_id: int) -> List[str]:
30
31
@pytest.mark.parametrize("tp_size", [4])
32
def test_mixtral_lora(mixtral_lora_files, tp_size):
33
+ """Original test, the LoRA model has the common target modules, not all"""
34
if torch.cuda.device_count() < tp_size:
35
pytest.skip(f"Not enough GPUs for tensor parallelism {tp_size}")
36
39
- llm = vllm.LLM(MODEL_PATH,
40
- enable_lora=True,
41
- max_num_seqs=16,
42
- max_loras=4,
43
- distributed_executor_backend="ray",
44
- tensor_parallel_size=tp_size)
37
+ prompts = [
38
+ "[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]", # noqa: E501
39
+ "[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]", # noqa: E501
40
+ "[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]", # noqa: E501
41
+ ]
42
+
43
+ llm = vllm.LLM(
44
+ MODEL_PATH,
45
+ enable_lora=True,
46
+ max_num_seqs=16,
47
+ max_loras=4,
48
+ distributed_executor_backend="ray",
49
+ tensor_parallel_size=tp_size,
50
+ )
51
52
expected_lora_output = [
53
"give_opinion(name[SpellForce 3], release_year[2017], developer[Grimlore Games], rating[poor])", # noqa: E501
54
"give_opinion(name[SpellForce 3], developer[Grimlore Games], release_year[2017], rating[poor])", # noqa: E501
55
"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])", # noqa: E501
56
]
51
- assert do_sample(llm, mixtral_lora_files,
52
- lora_id=1) == expected_lora_output
53
- assert do_sample(llm, mixtral_lora_files,
54
- lora_id=2) == expected_lora_output
57
+ assert do_sample(llm, mixtral_lora_files, lora_id=1,
58
+ prompts=prompts) == expected_lora_output
59
+ assert do_sample(llm, mixtral_lora_files, lora_id=2,
60
+ prompts=prompts) == expected_lora_output
61
+
62
+
63
+@pytest.mark.parametrize("tp_size", [4])
64
+def test_mixtral_lora_all_target_modules(mixtral_lora_files_all_target_modules,
65
+ tp_size):
66
+ """This LoRA model has all supported Mixtral target modules"""
67
+
68
+ if torch.cuda.device_count() < tp_size:
69
+ pytest.skip(f"Not enough GPUs for tensor parallelism {tp_size}")
70
+
71
+ prompts = [
72
+ "Question: What happens if you touch the eyes of a blind man?\nChoices:\nB: The man's sight will be restored.\nA: Nothing happens if you touch the eyes of a blind man.\nC: The man will get his eyesight back.\nD: The man will no longer be blind.\nAnswer:", # noqa: E501
73
+ "Question: Which will increase the energy of motion of water molecules?\nChoices:\nA: add heat\nB: add ice\nC: remove some of the water\nD: place the water in the freezer\nAnswer:", # noqa: E501
74
+ "Since Craig threw aluminum cans in the trash and Benjamin recycled, _ was environmentally irresponsible.\nChoices:\n1: Craig\n2: Benjamin\nAnswer:", # noqa: E501
75
+ ]
76
+
77
+ llm = vllm.LLM(
78
+ MODEL_PATH,
79
+ enable_lora=True,
80
+ max_num_seqs=16,
81
+ max_loras=4,
82
+ distributed_executor_backend="ray",
83
+ tensor_parallel_size=tp_size,
84
+ max_lora_rank=32,
85
+ )
86
+
87
+ expected_lora_output = [
88
+ "A: Nothing happens if you touch the eyes of a blind man.",
89
+ "A: add heat",
90
+ "1: Craig",
91
+ ]
92
+
93
+ assert do_sample(llm,
94
+ mixtral_lora_files_all_target_modules,
95
+ lora_id=1,
96
+ prompts=prompts) == expected_lora_output
97
+ assert do_sample(llm,
98
+ mixtral_lora_files_all_target_modules,
99
+ lora_id=2,
100
+ prompts=prompts) == expected_lora_output
vllm/model_executor/models/mixtral.py
+2
-4
@@ -322,10 +322,8 @@ class MixtralForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
322
323
# LoRA specific attributes
324
supported_lora_modules = [
325
- "qkv_proj",
326
- "o_proj",
327
- "embed_tokens",
328
- "lm_head",
325
+ "qkv_proj", "o_proj", "embed_tokens", "lm_head", "w1", "w2", "w3",
326
+ "gate"
327
]
328
embedding_modules = {
329
"embed_tokens": "input_embeddings",