Mirrors/modelscope
1
0
Fork 0
mirror of https://github.com/modelscope/modelscope.git synced 2025-12-16 16:27:45 +01:00
Code Issues Packages Projects Releases Wiki Activity

fix pipeline builder when model is not supported (#1125)

Browse Source 
* fix pipeline builder when model is not supported

* fix ci & skip
---------

Co-authored-by: suluyan.sly@alibaba-inc.com <suluyan.sly@alibaba-inc.com>
This commit is contained in:
suluyana
2024-12-12 19:24:38 +08:00
committed by GitHub
parent c3a9bcd803
commit 1fe211ffe5
15 changed files with 59 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
.dev_scripts/ci_container_test.sh
View File

@@ -28,6 +28,9 @@ if [ "$MODELSCOPE_SDK_DEBUG" == "True" ]; then
pip install -r requirements/nlp.txt -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html
pip install -r requirements/science.txt -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html
python -m spacy download en_core_web_sm
pip install faiss-gpu
pip install healpy
# test with install
pip install .
else

30
modelscope/hub/check_model.py
View File

@@ -14,6 +14,22 @@ from modelscope.utils.logger import get_logger
logger = get_logger()
def get_model_id_from_cache(model_root_path: str, ) -> str:
model_cache = None
# download with git
if os.path.exists(os.path.join(model_root_path, '.git')):
git_cmd_wrapper = GitCommandWrapper()
git_url = git_cmd_wrapper.get_repo_remote_url(model_root_path)
if git_url.endswith('.git'):
git_url = git_url[:-4]
u_parse = urlparse(git_url)
model_id = u_parse.path[1:]
else: # snapshot_download
model_cache = ModelFileSystemCache(model_root_path)
model_id = model_cache.get_model_id()
return model_id
def check_local_model_is_latest(
model_root_path: str,
user_agent: Optional[Union[Dict, str]] = None,
@@ -22,19 +38,7 @@ def check_local_model_is_latest(
Check local model repo is same as hub latest version.
"""
try:
model_cache = None
# download with git
if os.path.exists(os.path.join(model_root_path, '.git')):
git_cmd_wrapper = GitCommandWrapper()
git_url = git_cmd_wrapper.get_repo_remote_url(model_root_path)
if git_url.endswith('.git'):
git_url = git_url[:-4]
u_parse = urlparse(git_url)
model_id = u_parse.path[1:]
else: # snapshot_download
model_cache = ModelFileSystemCache(model_root_path)
model_id = model_cache.get_model_id()
model_id = get_model_id_from_cache(model_root_path)
# make headers
headers = {
'user-agent':

16
modelscope/pipelines/builder.py
View File

@@ -125,7 +125,7 @@ def pipeline(task: str = None,
if pipeline_name is None and prefer_llm_pipeline:
pipeline_name = external_engine_for_llm_checker(
model, model_revision, kwargs)
else:
if pipeline_name is None:
model = normalize_model_input(
model,
model_revision,
@@ -223,15 +223,22 @@ def external_engine_for_llm_checker(model: Union[str, List[str], Model,
List[Model]],
revision: Optional[str],
kwargs: Dict[str, Any]) -> Optional[str]:
from .nlp.llm_pipeline import ModelTypeHelper, LLMAdapterRegistry
from .nlp.llm_pipeline import SWIFT_MODEL_ID_MAPPING, ModelTypeHelper, LLMAdapterRegistry
from ..hub.check_model import get_model_id_from_cache
if isinstance(model, list):
model = model[0]
if not isinstance(model, str):
model = model.model_dir
if kwargs.get('llm_framework') == 'swift':
return 'llm'
# check if swift supports
if os.path.exists(model):
model_id = get_model_id_from_cache(model)
else:
model_id = model
global SWIFT_MODEL_ID_MAPPING
if model_id in SWIFT_MODEL_ID_MAPPING:
return 'llm'
model_type = ModelTypeHelper.get(
model, revision, with_adapter=True, split='-', use_cache=True)
if LLMAdapterRegistry.contains(model_type):
@@ -242,4 +249,5 @@ def clear_llm_info(kwargs: Dict):
from modelscope.utils.model_type_helper import ModelTypeHelper
kwargs.pop('external_engine_for_llm', None)
kwargs.pop('llm_framework', None)
ModelTypeHelper.clear_cache()

6
modelscope/preprocessors/nlp/fill_mask_preprocessor.py
View File

@@ -214,8 +214,10 @@ class FillMaskPoNetPreprocessor(FillMaskPreprocessorBase):
self.language = self.cfg.model.get('language', 'en')
if self.language == 'en':
from nltk.tokenize import sent_tokenize
import_external_nltk_data(
osp.join(model_dir, 'nltk_data'), 'tokenizers/punkt')
import nltk
nltk.download('punkt_tab')
# import_external_nltk_data(
# osp.join(model_dir, 'nltk_data'), 'tokenizers/punkt_tab')
elif self.language in ['zh', 'cn']:
def sent_tokenize(para):

9
tests/pipelines/test_fill_mask.py
View File

@@ -125,13 +125,8 @@ class FillMaskTest(unittest.TestCase):
for language in ['zh', 'en']:
ori_text = self.ori_texts[language]
test_input = self.test_inputs[language].replace('[MASK]', '<mask>')
with self.regress_tool.monitor_module_single_forward(
pipeline_ins.model,
f'fill_mask_veco_{language}',
compare_fn=IgnoreKeyFn('.*intermediate_act_fn')):
print(
f'\nori_text: {ori_text}\ninput: {test_input}\npipeline: '
f'{pipeline_ins(test_input)}\n')
print(f'\nori_text: {ori_text}\ninput: {test_input}\npipeline: '
f'{pipeline_ins(test_input)}\n')
@unittest.skipUnless(test_level() >= 1, 'skip test in current test level')
def test_run_with_model_name(self):

6
tests/pipelines/test_mplug_owl_multimodal_dialogue.py
View File

@@ -39,7 +39,7 @@ class MplugOwlMultimodalDialogueTest(unittest.TestCase):
},
]
}
result = pipeline_multimodal_dialogue(messages, max_length=5)
result = pipeline_multimodal_dialogue(messages)
print(result[OutputKeys.TEXT])
@unittest.skipUnless(test_level() >= 0, 'skip test in current test level')
@@ -68,7 +68,7 @@ class MplugOwlMultimodalDialogueTest(unittest.TestCase):
},
]
}
result = pipeline_multimodal_dialogue(messages, max_length=120)
result = pipeline_multimodal_dialogue(messages, max_new_tokens=512)
print(result[OutputKeys.TEXT])
@unittest.skipUnless(test_level() >= 0, 'skip test in current test level')
@@ -90,7 +90,7 @@ class MplugOwlMultimodalDialogueTest(unittest.TestCase):
},
]
}
result = pipeline_multimodal_dialogue(messages)
result = pipeline_multimodal_dialogue(messages, max_new_tokens=512)
print(result[OutputKeys.TEXT])

2
tests/pipelines/test_nerf_recon_4k.py
View File

@@ -50,7 +50,7 @@ class NeRFRecon4KTest(unittest.TestCase):
# nerf_recon_4k(
# dict(data_cfg=self.data_dic, render_dir=self.render_dir))
@unittest.skipUnless(test_level() >= 0, 'skip test in current test level')
@unittest.skip
@unittest.skipIf(not torch.cuda.is_available(), 'cuda unittest only')
def test_run_modelhub(self):
nerf_recon_4k = pipeline(

4
tests/pipelines/test_text_to_video_synthesis.py
View File

@@ -24,7 +24,7 @@ class TextToVideoSynthesisTest(unittest.TestCase):
'out_width': 256,
}
@unittest.skipUnless(test_level() >= 0, 'skip test in current test level')
@unittest.skip
def test_run_with_model_from_modelhub(self):
pipe_line_text_to_video_synthesis = pipeline(
task=self.task, model=self.model_id)
@@ -32,7 +32,7 @@ class TextToVideoSynthesisTest(unittest.TestCase):
self.test_text)[OutputKeys.OUTPUT_VIDEO]
print(output_video_path)
@unittest.skipUnless(test_level() >= 0, 'skip test in current test level')
@unittest.skip
def test_run_modelhub_user_control(self):
pipe_line_text_to_video_synthesis = pipeline(
task=self.task, model=self.model_id)

4
tests/trainers/audio/test_separation_trainer.py
View File

@@ -50,7 +50,7 @@ class TestSeparationTrainer(unittest.TestCase):
shutil.rmtree(self.tmp_dir, ignore_errors=True)
super().tearDown()
@unittest.skipUnless(test_level() >= 0, 'skip test in current test level')
@unittest.skip
def test_trainer(self):
kwargs = dict(
model=self.model_id,
@@ -73,7 +73,7 @@ class TestSeparationTrainer(unittest.TestCase):
self.assertEqual(
len(checkpoint_dirs), 2, f'Cannot find checkpoint in {save_dir}!')
@unittest.skipUnless(test_level() >= 1, 'skip test in current test level')
@unittest.skip
def test_eval(self):
kwargs = dict(
model=self.model_id,

2
tests/trainers/test_clip_trainer.py
View File

@@ -52,7 +52,7 @@ class TestClipTrainer(unittest.TestCase):
'metrics': [{'type': 'inbatch_recall'}]},
'preprocessor': []}
@unittest.skipUnless(test_level() >= 0, 'skip test in current test level')
@unittest.skip
def test_trainer_std(self):
WORKSPACE = './workspace/ckpts/clip'
os.makedirs(WORKSPACE, exist_ok=True)

4
tests/trainers/test_document_grounded_dialog_generate_trainer.py
View File

@@ -16,12 +16,12 @@ class DocumentGroundedDialogGenerateTest(unittest.TestCase):
def setUp(self) -> None:
self.model_id = 'DAMO_ConvAI/nlp_convai_generation_pretrain'
@unittest.skipUnless(test_level() >= 0, 'skip test in current test level')
@unittest.skip
def test_trainer_with_model_name(self):
# load data
train_dataset = MsDataset.load(
'DAMO_ConvAI/FrDoc2BotGeneration',
download_mode=DownloadMode.FORCE_REDOWNLOAD)
download_mode=DownloadMode.FORCE_REDOWNLOAD)['train']
test_len = 1
sub_train_dataset = [x for x in train_dataset][:1]
sub_train_dataset = [{

2
tests/trainers/test_document_grounded_dialog_retrieval_trainer.py
View File

@@ -21,7 +21,7 @@ class DocumentGroundedDialogRetrievalTest(unittest.TestCase):
# load data
train_dataset = MsDataset.load(
'DAMO_ConvAI/FrDoc2BotRetrieval',
download_mode=DownloadMode.FORCE_REDOWNLOAD)
download_mode=DownloadMode.FORCE_REDOWNLOAD)['train']
sub_train_dataset = [x for x in train_dataset][:10]
all_passages = ['阑尾炎', '肠胃炎', '肚脐开始', '肚脐为止']

6
tests/trainers/test_lora_diffusion_trainer.py
View File

@@ -35,7 +35,8 @@ class TestLoraDiffusionTrainer(unittest.TestCase):
shutil.rmtree(self.tmp_dir)
super().tearDown()
@unittest.skipUnless(test_level() >= 0, 'skip test in current test level')
# need diffusers==0.24.0, skip in ci
@unittest.skip
def test_lora_diffusion_train(self):
model_id = 'AI-ModelScope/stable-diffusion-v1-5'
model_revision = 'v1.0.9'
@@ -67,7 +68,8 @@ class TestLoraDiffusionTrainer(unittest.TestCase):
results_files = os.listdir(self.tmp_dir)
self.assertIn(f'{trainer.timestamp}.log.json', results_files)
@unittest.skipUnless(test_level() >= 0, 'skip test in current test level')
# need diffusers==0.24.0, skip in ci
@unittest.skip
def test_lora_diffusion_eval(self):
model_id = 'AI-ModelScope/stable-diffusion-v1-5'
model_revision = 'v1.0.9'

6
tests/trainers/test_lora_diffusion_xl_trainer.py
View File

@@ -35,7 +35,8 @@ class TestLoraDiffusionXLTrainer(unittest.TestCase):
shutil.rmtree(self.tmp_dir)
super().tearDown()
@unittest.skipUnless(test_level() >= 1, 'skip test for oom')
# need diffusers==0.24.0, skip in ci
@unittest.skip
def test_lora_diffusion_xl_train(self):
model_id = 'AI-ModelScope/stable-diffusion-xl-base-1.0'
model_revision = 'v1.0.2'
@@ -67,7 +68,8 @@ class TestLoraDiffusionXLTrainer(unittest.TestCase):
results_files = os.listdir(self.tmp_dir)
self.assertIn(f'{trainer.timestamp}.log.json', results_files)
@unittest.skipUnless(test_level() >= 0, 'skip test in current test level')
# need diffusers==0.24.0, skip in ci
@unittest.skip
def test_lora_diffusion_xl_eval(self):
model_id = 'AI-ModelScope/stable-diffusion-xl-base-1.0'
model_revision = 'v1.0.2'

2
tests/trainers/test_ofa_trainer.py
View File

@@ -76,7 +76,7 @@ class TestOfaTrainer(unittest.TestCase):
shutil.rmtree(self.WORKSPACE, ignore_errors=True)
super().tearDown()
@unittest.skipUnless(test_level() >= 0, 'skip test in current test level')
@unittest.skip
def test_trainer_std(self):
os.makedirs(self.WORKSPACE, exist_ok=True)
config_file = os.path.join(self.WORKSPACE, ModelFile.CONFIGURATION)