support ms-swift 3.0.0 (#1166)

.dev_scripts/ci_container_test.sh

@@ -32,6 +32,7 @@ if [ "$MODELSCOPE_SDK_DEBUG" == "True" ]; then
     pip install faiss-gpu
     pip install healpy
     pip install huggingface-hub==0.25.2
+    pip install ms-swift>=3.0.1
     # test with install
     pip install .
 else

modelscope/pipelines/builder.py

@@ -223,8 +223,9 @@ 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 SWIFT_MODEL_ID_MAPPING, init_swift_model_mapping, ModelTypeHelper, LLMAdapterRegistry
+    from .nlp.llm_pipeline import ModelTypeHelper, LLMAdapterRegistry
     from ..hub.check_model import get_model_id_from_cache
+    from swift.llm import get_model_info_meta
     if isinstance(model, list):
         model = model[0]
     if not isinstance(model, str):
@@ -237,9 +238,17 @@ def external_engine_for_llm_checker(model: Union[str, List[str], Model,
         else:
             model_id = model
 
-        init_swift_model_mapping()
-        if model_id.lower() in SWIFT_MODEL_ID_MAPPING:
+        try:
+            info = get_model_info_meta(model_id)
+            model_type = info[0].model_type
+        except Exception as e:
+            logger.warning(
+                f'Cannot using llm_framework with {model_id}, '
+                f'ignoring llm_framework={self.llm_framework} : {e}')
+            model_type = None
+        if model_type:
             return 'llm'
+
     model_type = ModelTypeHelper.get(
         model, revision, with_adapter=True, split='-', use_cache=True)
     if LLMAdapterRegistry.contains(model_type):

modelscope/pipelines/nlp/llm_pipeline.py

@@ -29,21 +29,9 @@ from modelscope.utils.streaming_output import (PipelineStreamingOutputMixin,
 
 logger = get_logger()
 
-SWIFT_MODEL_ID_MAPPING = {}
 SWIFT_FRAMEWORK = 'swift'
 
 
-def init_swift_model_mapping():
-    from swift.llm.utils import MODEL_MAPPING
-
-    global SWIFT_MODEL_ID_MAPPING
-    if not SWIFT_MODEL_ID_MAPPING:
-        SWIFT_MODEL_ID_MAPPING = {
-            v['model_id_or_path'].lower(): k
-            for k, v in MODEL_MAPPING.items()
-        }
-
-
 class LLMAdapterRegistry:
 
     llm_format_map = {'qwen': [None, None, None]}
@@ -227,12 +215,12 @@ class LLMPipeline(Pipeline, PipelineStreamingOutputMixin):
 
     def _init_swift(self, model_id, device) -> None:
         from swift.llm import prepare_model_template
-        from swift.llm.utils import InferArguments
+        from swift.llm import InferArguments, get_model_info_meta
 
         def format_messages(messages: Dict[str, List[Dict[str, str]]],
                             tokenizer: PreTrainedTokenizer,
                             **kwargs) -> Dict[str, torch.Tensor]:
-            inputs, _ = self.template.encode(get_example(messages))
+            inputs = self.template.encode(messages)
             inputs.pop('labels', None)
             if 'input_ids' in inputs:
                 input_ids = torch.tensor(inputs['input_ids'])[None]
@@ -265,12 +253,7 @@ class LLMPipeline(Pipeline, PipelineStreamingOutputMixin):
             else:
                 return dict(system=system, prompt=prompt, history=history)
 
-        init_swift_model_mapping()
-
-        assert model_id.lower() in SWIFT_MODEL_ID_MAPPING,\
-            f'Invalid model id {model_id} or Swift framework does not support this model.'
-        args = InferArguments(
-            model_type=SWIFT_MODEL_ID_MAPPING[model_id.lower()])
+        args = InferArguments(model=model_id)
         model, template = prepare_model_template(
             args, device_map=self.device_map)
         self.model = add_stream_generate(model)