Add Ola Model

vlmeval/config.py

@@ -458,6 +458,10 @@
     'valley_eagle': partial(ValleyEagleChat, model_path='bytedance-research/Valley-Eagle-7B'),
 }
 
+ola_series = {
+    'ola': partial(Ola, model_path='THUdyh/Ola-7b'),
+}
+
 ross_series = {
     'ross-qwen2-7b': partial(Ross, model_path='HaochenWang/ross-qwen2-7b'),
 }
@@ -477,7 +481,7 @@
     mantis_series, mmalaya_series, phi3_series, xgen_mm_series, qwen2vl_series,
     slime_series, eagle_series, moondream_series, llama_series, molmo_series,
     kosmos_series, points_series, nvlm_series, vintern_series, h2ovl_series, aria_series,
-    smolvlm_series, sail_series, valley_series, vita_series, ross_series, emu_series, ursa_series
+    smolvlm_series, sail_series, valley_series, vita_series, ross_series, emu_series, ola_series, ursa_series
 ]
 
 for grp in model_groups:

vlmeval/vlm/__init__.py

@@ -64,4 +64,5 @@
 from .sail_vl import SailVL
 from .valley import ValleyEagleChat
 from .ross import Ross
-from .ursa import UrsaChat
+from .ola import Ola
+from .ursa import UrsaChat

vlmeval/vlm/ola/__init__.py

@@ -0,0 +1 @@
+from .ola_model import Ola

vlmeval/vlm/ola/ola/arguments.py

@@ -0,0 +1,65 @@
+import transformers
+
+from dataclasses import dataclass, field
+from typing import Optional
+
+
+@dataclass
+class ModelArguments:
+    model_name_or_path: Optional[str] = field(default="facebook/opt-125m")
+    version: Optional[str] = field(default="v0")
+    freeze_backbone: bool = field(default=False)
+    tune_speech_projector: bool = field(default=False)
+    tune_speech_encoder: bool = field(default=False)
+    tune_speech_generator_only: bool = field(default=False)
+    speech_encoder_type: Optional[str] = field(default=None)
+    speech_encoder: Optional[str] = field(default=None)
+    pretrain_speech_projector: Optional[str] = field(default=None)
+    speech_projector_type: Optional[str] = field(default='linear')
+    speech_encoder_ds_rate: int = 5
+    speech_encoder_hidden_size: int = 1280
+
+
+@dataclass
+class DataArguments:
+    data_path: str = field(default=None,
+                           metadata={"help": "Path to the training data."})
+    is_multimodal: bool = False
+    input_type: str = field(default="mel")
+    speech_normalize: bool = False
+    mel_size: int = 128
+    has_tgt_units: bool = False
+
+
+@dataclass
+class TrainingArguments(transformers.TrainingArguments):
+    cache_dir: Optional[str] = field(default=None)
+    optim: str = field(default="adamw_torch")
+    freeze_speech_projector: bool = field(default=False)
+    model_max_length: int = field(
+        default=512,
+        metadata={
+            "help":
+            "Maximum sequence length. Sequences will be right padded (and possibly truncated)."
+        },
+    )
+    double_quant: bool = field(
+        default=True,
+        metadata={"help": "Compress the quantization statistics through double quantization."}
+    )
+    quant_type: str = field(
+        default="nf4",
+        metadata={"help": "Quantization data type to use. Should be one of `fp4` or `nf4`."}
+    )
+    bits: int = field(
+        default=16,
+        metadata={"help": "How many bits to use."}
+    )
+    lora_enable: bool = False
+    lora_r: int = 64
+    lora_alpha: int = 16
+    lora_dropout: float = 0.05
+    lora_weight_path: str = ""
+    lora_bias: str = "none"
+    speech_projector_lr: Optional[float] = None
+    group_by_modality_length: bool = field(default=False)

vlmeval/vlm/ola/ola/constants.py

@@ -0,0 +1,14 @@
+CONTROLLER_HEART_BEAT_EXPIRATION = 30
+WORKER_HEART_BEAT_INTERVAL = 15
+
+LOGDIR = "."
+
+# Model Constants
+IGNORE_INDEX = -100
+SPEECH_TOKEN_INDEX = -200
+DEFAULT_SPEECH_TOKEN = "<speech>"
+IMAGE_TOKEN_INDEX= -300
+DEFAULT_IMAGE_TOKEN = "<image>"
+DEFAULT_IMAGE_PATCH_TOKEN = "<im_patch>"
+DEFAULT_IM_START_TOKEN = "<im_start>"
+DEFAULT_IM_END_TOKEN = "<im_end>"

vlmeval/vlm/ola/ola/conversation.py

@@ -0,0 +1,137 @@
+import dataclasses
+from enum import auto, Enum
+from typing import List, Any, Union, Tuple
+import base64
+from io import BytesIO
+from PIL import Image
+
+
+class SeparatorStyle(Enum):
+    """Different separator style."""
+    TWO = auto()
+    PLAIN = auto()
+    CHATML = auto()
+    LLAMA_2 = auto()
+    LLAMA_3 = auto()
+    QWEN2 = auto()
+
+
+@dataclasses.dataclass
+class Conversation:
+    """A class that keeps all conversation history."""
+    system: str
+    roles: List[str]
+    messages: List[List[str]]
+    offset: int
+    sep_style: SeparatorStyle = SeparatorStyle.PLAIN
+    sep: str = "###"
+    sep2: str = None
+    version: str = "Unknown"
+
+    tokenizer_id: str = ""
+    tokenizer: Any = None
+    # Stop criteria (the default one is EOS token)
+    stop_str: Union[str, List[str]] = None
+    # Stops generation if meeting any token in this list
+    stop_token_ids: List[int] = None
+
+    skip_next: bool = False
+
+    def get_prompt(self):
+        messages = self.messages
+
+        if self.sep_style == SeparatorStyle.TWO:
+            seps = [self.sep, self.sep2]
+            ret = self.system + seps[0]
+            for i, (role, message) in enumerate(messages):
+                if message:
+                    if type(message) is tuple:
+                        message = message[0]
+                    ret += role + ": " + message + seps[i % 2]
+                else:
+                    ret += role + ":"
+        elif self.sep_style == SeparatorStyle.QWEN2:
+            start = '<|im_start|>'
+            end = '<|im_end|>\n'
+            ret = start + 'system\n' + self.system + end
+            for i, (role, message) in enumerate(messages):
+                if message:
+                    if type(message) is tuple:
+                        message, _, _ = message
+
+                    if message.endswith('<|endoftext|>'):
+                        message = message.replace('<|endoftext|>', '')
+                        ret += start + role + "\n" + message + end + '<|endoftext|>'                        
+                    else:
+                        assert not '<|endoftext|>' in message, f"Invalid message: {message}"
+                        ret += start + role + "\n" + message + end
+                else:
+                    ret += start + role + "\n"
+        else:
+            raise ValueError(f"Invalid style: {self.sep_style}")
+
+        return ret
+
+    def append_message(self, role, message):
+        self.messages.append([role, message])
+
+    def to_gradio_chatbot(self):
+        ret = []
+        for i, (role, msg) in enumerate(self.messages[self.offset:]):
+            if i % 2 == 0:
+                if type(msg) is tuple:
+                    msg, speech = msg
+                    ret.append([msg, None])
+                else:
+                    ret.append([msg, None])
+            else:
+                ret[-1][-1] = msg
+        return ret
+
+    def copy(self):
+        return Conversation(
+            system=self.system,
+            roles=self.roles,
+            messages=[[x, y] for x, y in self.messages],
+            offset=self.offset,
+            sep_style=self.sep_style,
+            sep=self.sep,
+            sep2=self.sep2,
+            version=self.version)
+
+    def dict(self):
+        if len(self.get_images()) > 0:
+            return {
+                "system": self.system,
+                "roles": self.roles,
+                "messages": [[x, y[0] if type(y) is tuple else y] for x, y in self.messages],
+                "offset": self.offset,
+                "sep": self.sep,
+                "sep2": self.sep2,
+            }
+        return {
+            "system": self.system,
+            "roles": self.roles,
+            "messages": self.messages,
+            "offset": self.offset,
+            "sep": self.sep,
+            "sep2": self.sep2,
+        }
+
+conv_qwen_v1 = Conversation(
+    system="You are a helpful assistant.",
+    roles=("user", "assistant"),
+    version="v1",
+    messages=(),
+    offset=0,
+    sep_style=SeparatorStyle.QWEN2,
+)
+
+default_conversation = conv_qwen_v1
+conv_templates = {
+    'v1_qwen2': conv_qwen_v1,
+}
+
+
+if __name__ == "__main__":
+    print(default_conversation.get_prompt())