fix glm4v (#2806)

swift/llm/__init__.py

@@ -20,7 +20,7 @@
                         HfConfigFactory, ModelInfo, ModelMeta, ModelKeys, register_model_arch, MultiModelKeys,
                         ModelArch, get_model_arch, MODEL_ARCH_MAPPING, get_model_info_meta, get_model_name, ModelGroup,
                         Model, get_model_tokenizer_with_flash_attn, get_model_tokenizer_multimodal, load_by_unsloth,
-                        git_clone_github)
+                        git_clone_github, get_matched_model_meta)
     from .dataset import (AlpacaPreprocessor, ResponsePreprocessor, MessagesPreprocessor, AutoPreprocessor,
                           DATASET_MAPPING, MediaResource, register_dataset, register_dataset_info, EncodePreprocessor,
                           LazyLLMDataset, ConstantLengthDataset, standard_keys, load_dataset, DATASET_TYPE,
@@ -54,7 +54,7 @@
             'ModelInfo', 'ModelMeta', 'ModelKeys', 'register_model_arch', 'MultiModelKeys', 'ModelArch',
             'MODEL_ARCH_MAPPING', 'get_model_arch', 'get_model_info_meta', 'get_model_name', 'register_model',
             'ModelGroup', 'Model', 'get_model_tokenizer_with_flash_attn', 'get_model_tokenizer_multimodal',
-            'load_by_unsloth', 'git_clone_github'
+            'load_by_unsloth', 'git_clone_github', 'get_matched_model_meta'
         ],
         'dataset': [
             'AlpacaPreprocessor', 'ClsPreprocessor', 'ComposePreprocessor', 'MessagesPreprocessor', 'DATASET_MAPPING',

swift/llm/argument/rlhf_args.py

@@ -48,7 +48,7 @@ def __post_init__(self):
         self._set_default()
         super().__post_init__()
 
-        if self.rlhf_type not in ['cpo', 'orpo', 'rm'] and (self.train_type == 'full' or self.rlhf_type == 'ppo'):
+        if self.rlhf_type in ['dpo', 'kto'] and self.train_type == 'full' or self.rlhf_type == 'ppo':
             self.ref_model = self.ref_model or self.model
             self.ref_model_type = self.ref_model_type or self.model_type
             self.ref_model_revision = self.ref_model_revision or self.model_revision

swift/llm/model/__init__.py

@@ -5,5 +5,5 @@
 from .register import (MODEL_MAPPING, Model, ModelGroup, ModelMeta, fix_do_sample_warning, get_default_device_map,
                        get_default_torch_dtype, get_matched_model_meta, get_model_info_meta, get_model_name,
                        get_model_tokenizer, get_model_tokenizer_multimodal, get_model_tokenizer_with_flash_attn,
-                       get_model_with_value_head, load_by_unsloth, register_model)
+                       load_by_unsloth, register_model)
 from .utils import HfConfigFactory, ModelInfo, git_clone_github, safe_snapshot_download

swift/llm/model/constant.py

@@ -93,7 +93,7 @@ class LLMModelType:
     mamba = 'mamba'
     polylm = 'polylm'
     aya = 'aya'
-
+    # bert
     modern_bert = 'modern_bert'
     bert = 'bert'
 

swift/llm/model/model/glm.py

@@ -178,14 +178,18 @@ def get_model_tokenizer_glm4v(model_dir: str,
 register_model(
     ModelMeta(
         MLLMModelType.glm4v,
-        [ModelGroup([
-            Model('ZhipuAI/glm-4v-9b', 'THUDM/glm-4v-9b'),
-        ])],
+        [
+            ModelGroup(
+                [
+                    Model('ZhipuAI/glm-4v-9b', 'THUDM/glm-4v-9b'),
+                ],
+                requires=['transformers>=4.42,<4.45'],
+            ),
+        ],
         TemplateType.glm4v,
         get_model_tokenizer_glm4v,
         architectures=['ChatGLMModel', 'ChatGLMForConditionalGeneration'],
         model_arch=ModelArch.glm4v,
-        requires=['transformers>=4.42'],
     ))
 
 

swift/llm/model/register.py

@@ -196,62 +196,6 @@ def get_model_tokenizer_from_local(model_dir: str,
     return model, tokenizer
 
 
-def get_model_with_value_head(model) -> 'AutoModelForCausalLMWithValueHead':
-    from trl import AutoModelForCausalLMWithValueHead
-    lm_head_namings = ['lm_head', 'embed_out']
-    if not any(hasattr(model, attribute) for attribute in lm_head_namings):
-        setattr(model, 'lm_head', None)  # avoid ValueError
-
-    model = AutoModelForCausalLMWithValueHead.from_pretrained(model)
-
-    def patch_valuehead_model(model):
-        attr_list = [
-            'get_input_embeddings', 'vis_processor', 'extract_feature', 'get_rope_index', 'model', 'vision_tower',
-            'img2emb', '_encode_image', '_merge_input_ids_with_image_features', 'prepare_inputs_embeds',
-            'build_conversation_input_ids', 'config', 'get_slice_image_placeholder', 'transform', 'get_vllm_embedding',
-            'forward_image', 'dtype', 'base_model_prefix', 'device', 'visual'
-        ]
-        for attr in attr_list:
-            if hasattr(model.pretrained_model, attr) and not hasattr(model, attr):
-                setattr(model, attr, getattr(model.pretrained_model, attr))
-
-        # PPO compatible
-        if not hasattr(model, 'score'):
-            setattr(model, 'score', model.v_head)
-        if model.base_model_prefix == '' and hasattr(model.pretrained_model, 'language_model'):
-            model.base_model_prefix = model.pretrained_model.language_model.base_model_prefix
-
-        base_model_prefix = model.pretrained_model.base_model_prefix
-        if hasattr(model.pretrained_model, base_model_prefix):
-            setattr(model, base_model_prefix, getattr(model.pretrained_model, base_model_prefix))
-
-    patch_valuehead_model(model)
-
-    # try to load local vhead weights
-    vhead_params = None
-    try:
-        from safetensors import safe_open
-        vhead_file = os.path.join(model.pretrained_model.model_dir, 'value_head.safetensors')
-        with safe_open(vhead_file, framework='pt', device='cpu') as f:
-            vhead_params = {key: f.get_tensor(key) for key in f.keys()}
-    except Exception:
-        pass
-
-    try:
-        vhead_file = os.path.join(model.pretrained_model.model_dir, 'value_head.bin')
-        vhead_params = torch.load(vhead_file, map_location='cpu')
-    except Exception:
-        pass
-
-    if vhead_params is not None:
-        model.load_state_dict(vhead_params, strict=False)
-        logger.info(f'Loading value head weights from {vhead_file}')
-    else:
-        logger.info('The local value head weight file was not detected.'
-                    'Ignore it if this is during the reward modeling phase,')
-    return model
-
-
 def get_model_tokenizer_with_flash_attn(model_dir: str,
                                         model_info: ModelInfo,
                                         model_kwargs: Dict[str, Any],
@@ -430,7 +374,7 @@ def get_model_info_meta(
     if model_meta is None and model_type is not None:
         model_meta = MODEL_MAPPING[model_type]
     if model_meta is None:
-        model_meta = ModelMeta('', [], 'dummy', get_model_tokenizer_from_local, model_arch=None)
+        model_meta = ModelMeta(None, [], 'dummy', get_model_tokenizer_from_local, model_arch=None)
         logger.info(f'Temporarily create model_meta: {model_meta}')
 
     if torch_dtype is None:

swift/llm/template/template/glm.py

@@ -64,6 +64,7 @@ def _encode(self, inputs: StdTemplateInputs) -> Dict[str, Any]:
             encoded['images'] = inputs2['images']
         encoded['input_ids'] = input_ids
         encoded['labels'] = labels
+        encoded['position_ids'] = list(range(len(input_ids)))
         return encoded
 
     def _data_collator(self, batch: List[Dict[str, Any]], *, padding_to: Optional[int] = None) -> Dict[str, Any]:

swift/llm/template/template_inputs.py

@@ -45,7 +45,7 @@ def __post_init__(self):
             val = getattr(self, key)
             if isinstance(val, str):
                 setattr(self, key, [val])
-
+        assert isinstance(self.messages, list), f'messages: {self.messages}'
         self.remove_response(self.messages)
 
     @staticmethod

swift/ui/base.py

@@ -15,8 +15,7 @@
 from gradio import Accordion, Audio, Button, Checkbox, Dropdown, File, Image, Slider, Tab, TabItem, Textbox, Video
 from modelscope.hub.utils.utils import get_cache_dir
 
-from swift.llm import TEMPLATE_MAPPING, BaseArguments
-from swift.llm.model.register import get_matched_model_meta
+from swift.llm import TEMPLATE_MAPPING, BaseArguments, get_matched_model_meta
 
 all_langs = ['zh', 'en']
 builder: Type['BaseUI'] = None

tests/test_align/test_template/test_vision.py

@@ -17,6 +17,8 @@ def _infer_model(pt_engine, system=None, messages=None, images=None):
         resp = pt_engine.infer([{'messages': messages}], request_config=request_config)
         response = resp[0].choices[0].message.content
         messages += [{'role': 'assistant', 'content': response}, {'role': 'user', 'content': '<image>这是什么'}]
+    else:
+        messages = messages.copy()
     if images is None:
         images = ['http://modelscope-open.oss-cn-hangzhou.aliyuncs.com/images/cat.png']
     resp = pt_engine.infer([{'messages': messages, 'images': images}], request_config=request_config)
@@ -69,9 +71,14 @@ def test_yi_vl():
 def test_glm4v():
     # There will be differences in '\n'. This is normal.
     pt_engine = PtEngine('ZhipuAI/glm-4v-9b')
-    _infer_model(pt_engine)
+    messages = [{'role': 'user', 'content': '描述这张图片'}]
+    response = _infer_model(pt_engine, messages=messages)
     pt_engine.default_template.template_backend = 'jinja'
-    _infer_model(pt_engine)
+    response2 = _infer_model(pt_engine, messages=messages)
+    assert response == ('这张图片是一只小猫的特写，它有着非常醒目的蓝色眼睛和混合了灰色、白色和棕色毛发的皮毛。小猫的耳朵竖立着，胡须清晰可见。它的眼神看起来既好奇又警觉，整体上显得非常可爱。')
+    assert response2 == ('这是一张特写照片，展示了一只毛茸茸的小猫。小猫的眼睛大而圆，呈深蓝色，眼珠呈金黄色，非常明亮。它的鼻子短而小巧，'
+                         '是粉色的。小猫的嘴巴紧闭，胡须细长。它的耳朵竖立着，耳朵内侧是白色的，外侧是棕色的。小猫的毛发看起来柔软而浓密，'
+                         '主要是白色和棕色相间的花纹。背景模糊不清，但似乎是一个室内环境。')
 
 
 def test_minicpmv():
@@ -307,11 +314,11 @@ def test_doc_owl2():
     # test_deepseek_vl()
     # test_deepseek_vl2()
     # test_qwen_vl()
-    # test_glm4v()
+    test_glm4v()
     # test_minicpmv()
     # test_got_ocr()
     # test_paligemma()
-    test_paligemma2()
+    # test_paligemma2()
     # test_pixtral()
     # test_llama_vision()
     # test_llava_hf()