Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[FIX] hymba quant needs desc_act=False #710

Merged
merged 1 commit into from
Nov 29, 2024
Merged

[FIX] hymba quant needs desc_act=False #710

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions tests/models/model_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,6 +36,9 @@ class ModelTest(unittest.TestCase):
MODEL_MAX_LEN = 4096
DELETE_QUANTIZED_MODEL = True

# quant config
DESC_ACT = True

def generate(self, model, tokenizer, prompt=None):
if prompt is None:
prompt = "I am in Paris and"
Expand Down Expand Up @@ -83,6 +86,7 @@ def quantModel(self, model_id_or_path, trust_remote_code=False, torch_dtype="aut
bits=4,
group_size=128,
format=FORMAT.GPTQ,
desc_act=self.DESC_ACT,
)
model = GPTQModel.load(
model_id_or_path,
Expand Down
8 changes: 6 additions & 2 deletions tests/models/test_hymba.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,14 +4,18 @@

class TestHymba(ModelTest):
NATIVE_MODEL_ID = "/monster/data/model/Hymba-1.5B-Instruct/" # "baichuan-inc/Baichuan2-7B-Chat"
NATIVE_ARC_CHALLENGE_ACC = 0.2073
NATIVE_ARC_CHALLENGE_ACC_NORM = 0.2713
NATIVE_ARC_CHALLENGE_ACC = 0.3814
NATIVE_ARC_CHALLENGE_ACC_NORM = 0.3686
MODEL_MAX_LEN = 8192
TRUST_REMOTE_CODE = True
APPLY_CHAT_TEMPLATE = True
# Hymba currently only supports a batch size of 1.
# See https://huggingface.co/nvidia/Hymba-1.5B-Instruct
BATCH_SIZE = 1

# Hymba currently tests that DESC_ACT=False to get better results.
# If DESC_ACT=False, the output will be terrible.
DESC_ACT = False

def test_hymba(self):
self.quant_lm_eval()