Merge branch 'main' of https://github.com/hpcaitech/ColossalAI

applications/Chat/coati/models/base/reward_model.py

@@ -17,11 +17,13 @@ class RewardModel(LoRAModule):
         lora_train_bias (str): LoRA bias training mode.
     """
 
-    def __init__(self,
+    def __init__(
+        self,
         model: nn.Module,
         value_head: Optional[nn.Module] = None,
         lora_rank: int = 0,
-                 lora_train_bias: str = 'none') -> None:
+        lora_train_bias: str = "none",
+    ) -> None:
         super().__init__(lora_rank=lora_rank, lora_train_bias=lora_train_bias)
         self.model = model
         self.convert_to_lora()
@@ -33,9 +35,12 @@ class RewardModel(LoRAModule):
         else:
             self.value_head = nn.Linear(model.config.n_embd, 1)
 
-    def forward(self, sequences: torch.LongTensor, attention_mask: Optional[torch.Tensor] = None) -> torch.Tensor:
+    def forward(self, sequences: torch.LongTensor, attention_mask: torch.Tensor) -> torch.Tensor:
         outputs = self.model(sequences, attention_mask=attention_mask)
-        last_hidden_states = outputs['last_hidden_state']
-        values = self.value_head(last_hidden_states)[:, :-1]
-        value = values.mean(dim=1).squeeze(1)    # ensure shape is (B)
-        return value
+        last_hidden_states = outputs["last_hidden_state"]
+        sequence_lengths = torch.max(attention_mask * torch.arange(sequences.size(1), device=sequences.device), dim=1)[
+            0
+        ]
+        sequence_hidden_states = last_hidden_states[torch.arange(last_hidden_states.size(0)), sequence_lengths]
+        values = self.value_head(sequence_hidden_states).squeeze(1)  # ensure shape is (B, )
+        return values

applications/Chat/coati/models/bloom/__init__.py

@@ -2,4 +2,4 @@ from .bloom_actor import BLOOMActor
 from .bloom_critic import BLOOMCritic
 from .bloom_rm import BLOOMRM
 
-__all__ = ['BLOOMActor', 'BLOOMCritic', 'BLOOMRM']
+__all__ = ["BLOOMActor", "BLOOMCritic", "BLOOMRM"]

applications/Chat/coati/models/bloom/bloom_actor.py

 from typing import Optional
 
-import torch
-from transformers import BloomConfig, BloomForCausalLM, BloomModel
+from transformers import BloomConfig, BloomForCausalLM
 
 from ..base import Actor
 
@@ -18,12 +17,14 @@ class BLOOMActor(Actor):
         lora_train_bias (str): LoRA bias training mode.
     """
 
-    def __init__(self,
+    def __init__(
+        self,
         pretrained: str = None,
         config: Optional[BloomConfig] = None,
         checkpoint: bool = False,
         lora_rank: int = 0,
-                 lora_train_bias: str = 'none') -> None:
+        lora_train_bias: str = "none",
+    ) -> None:
         if pretrained is not None:
             model = BloomForCausalLM.from_pretrained(pretrained)
         elif config is not None:

applications/Chat/coati/models/bloom/bloom_critic.py

 from typing import Optional
 
-import torch
 import torch.nn as nn
-from transformers import BloomConfig, BloomForCausalLM, BloomModel
+from transformers import BloomConfig, BloomModel
 
 from ..base import Critic
 
@@ -14,25 +13,24 @@ class BLOOMCritic(Critic):
     Args:
         pretrained (str): Pretrained model name or path.
         config (BloomConfig): Model config.
-        checkpoint (bool): Enable gradient checkpointing.
         lora_rank (int): LoRA rank.
         lora_train_bias (str): LoRA bias training mode.
     """
 
-    def __init__(self,
+    def __init__(
+        self,
         pretrained: str = None,
         config: Optional[BloomConfig] = None,
-                 checkpoint: bool = False,
         lora_rank: int = 0,
-                 lora_train_bias: str = 'none',
-                 **kwargs) -> None:
+        lora_train_bias: str = "none",
+        **kwargs,
+    ) -> None:
         if pretrained is not None:
             model = BloomModel.from_pretrained(pretrained)
         elif config is not None:
             model = BloomModel(config)
         else:
             model = BloomModel(BloomConfig())
-        if checkpoint:
-            model.gradient_checkpointing_enable()
+
         value_head = nn.Linear(model.config.hidden_size, 1)
         super().__init__(model, value_head, lora_rank, lora_train_bias, **kwargs)

applications/Chat/coati/models/bloom/bloom_rm.py

 from typing import Optional
 
 import torch.nn as nn
-from transformers import BloomConfig, BloomForCausalLM, BloomModel
+from transformers import BloomConfig, BloomModel
 
 from ..base import RewardModel
 
@@ -13,25 +13,24 @@ class BLOOMRM(RewardModel):
     Args:
         pretrained (str): Pretrained model name or path.
         config (BloomConfig): Model config.
-        checkpoint (bool): Enable gradient checkpointing.
         lora_rank (int): LoRA rank.
         lora_train_bias (str): LoRA bias training mode.
     """
 
-    def __init__(self,
+    def __init__(
+        self,
         pretrained: str = None,
         config: Optional[BloomConfig] = None,
-                 checkpoint: bool = False,
         lora_rank: int = 0,
-                 lora_train_bias: str = 'none') -> None:
+        lora_train_bias: str = "none",
+    ) -> None:
         if pretrained is not None:
             model = BloomModel.from_pretrained(pretrained)
         elif config is not None:
             model = BloomModel(config)
         else:
             model = BloomModel(BloomConfig())
-        if checkpoint:
-            model.gradient_checkpointing_enable()
+
         value_head = nn.Linear(model.config.hidden_size, 1)
         value_head.weight.data.normal_(mean=0.0, std=1 / (model.config.hidden_size + 1))
         super().__init__(model, value_head, lora_rank, lora_train_bias)

applications/Chat/coati/models/chatglm/__init__.py

+from .chatglm_actor import ChatGLMActor
+
+__all__ = ["ChatGLMActor"]

applications/Chat/coati/models/chatglm/chatglm_actor.py

+from typing import Optional
+
+from ..base import Actor
+from .configuration_chatglm import ChatGLMConfig
+from .modeling_chatglm import ChatGLMForConditionalGeneration
+
+
+class ChatGLMActor(Actor):
+    """
+    ChatGLM Actor model.
+
+    Args:
+        pretrained (str): Pretrained model name or path.
+        config (ChatGLMConfig): Model config.
+        checkpoint (bool): Enable gradient checkpointing.
+
+    do not support lora for now.
+    """
+
+    def __init__(
+        self, pretrained: str = None, config: Optional[ChatGLMConfig] = None, checkpoint: bool = False
+    ) -> None:
+        if pretrained is not None:
+            model = ChatGLMForConditionalGeneration.from_pretrained(pretrained)
+        elif config is not None:
+            model = ChatGLMForConditionalGeneration(config)
+        else:
+            model = ChatGLMForConditionalGeneration(ChatGLMConfig())
+        if checkpoint:
+            model.gradient_checkpointing_enable()
+        super().__init__(model, lora_rank=0, lora_train_bias="none")

applications/Chat/coati/models/chatglm/chatglm_tokenizer.py

+"""
+This code is copied from https://huggingface.co/THUDM/chatglm-6b/blob/main/tokenization_chatglm.py
+"""
+"""Tokenization classes for ChatGLM."""
+import os
+from typing import Dict, List, Optional, Union
+
+import numpy as np
+import sentencepiece as spm
+from transformers.tokenization_utils import PreTrainedTokenizer
+from transformers.tokenization_utils_base import BatchEncoding, EncodedInput
+from transformers.utils import PaddingStrategy, logging
+
+logger = logging.get_logger(__name__)
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    "THUDM/chatglm-6b": 2048,
+}
+
+
+class TextTokenizer:
+    def __init__(self, model_path):
+        self.sp = spm.SentencePieceProcessor()
+        self.sp.Load(model_path)
+        self.num_tokens = self.sp.vocab_size()
+
+    def encode(self, text):
+        return self.sp.EncodeAsIds(text)
+
+    def decode(self, ids: List[int]):
+        return self.sp.DecodeIds(ids)
+
+    def tokenize(self, text):
+        return self.sp.EncodeAsPieces(text)
+
+    def convert_tokens_to_string(self, tokens):
+        return self.sp.DecodePieces(tokens)
+
+    def convert_tokens_to_ids(self, tokens):
+        return [self.sp.PieceToId(token) for token in tokens]
+
+    def convert_token_to_id(self, token):
+        return self.sp.PieceToId(token)
+
+    def convert_id_to_token(self, idx):
+        return self.sp.IdToPiece(idx)
+
+    def __len__(self):
+        return self.num_tokens
+
+
+class SPTokenizer:
+    def __init__(
+        self,
+        vocab_file,
+        num_image_tokens=20000,
+        max_blank_length=80,
+        byte_fallback=True,
+    ):
+        assert vocab_file is not None
+        self.vocab_file = vocab_file
+        self.num_image_tokens = num_image_tokens
+        self.special_tokens = ["[MASK]", "[gMASK]", "[sMASK]", "<unused_0>", "<sop>", "<eop>", "<ENC>", "<dBLOCK>"]
+        self.max_blank_length = max_blank_length
+        self.byte_fallback = byte_fallback
+        self.text_tokenizer = TextTokenizer(vocab_file)
+
+    def _get_text_tokenizer(self):
+        return self.text_tokenizer
+
+    @staticmethod
+    def get_blank_token(length: int):
+        assert length >= 2
+        return f"<|blank_{length}|>"
+
+    @staticmethod
+    def get_tab_token():
+        return f"<|tab|>"
+
+    @property
+    def num_text_tokens(self):
+        return self.text_tokenizer.num_tokens
+
+    @property
+    def num_tokens(self):
+        return self.num_image_tokens + self.num_text_tokens
+
+    @staticmethod
+    def _encode_whitespaces(text: str, max_len: int = 80):
+        text = text.replace("\t", SPTokenizer.get_tab_token())
+        for i in range(max_len, 1, -1):
+            text = text.replace(" " * i, SPTokenizer.get_blank_token(i))
+        return text
+
+    def _preprocess(self, text: str, linebreak=True, whitespaces=True):
+        if linebreak:
+            text = text.replace("\n", "<n>")
+        if whitespaces:
+            text = self._encode_whitespaces(text, max_len=self.max_blank_length)
+        return text
+
+    def encode(self, text: str, linebreak=True, whitespaces=True, add_dummy_prefix=True) -> List[int]:
+        """
+        @param text: Text to encode.
+        @param linebreak: Whether to encode newline (\n) in text.
+        @param whitespaces: Whether to encode multiple whitespaces or tab in text, useful for source code encoding.
+        @param special_tokens: Whether to encode special token ([MASK], [gMASK], etc.) in text.
+        @param add_dummy_prefix: Whether to add dummy blank space in the beginning.
+        """
+        text = self._preprocess(text, linebreak, whitespaces)
+        if not add_dummy_prefix:
+            text = "<n>" + text
+        tmp = self._get_text_tokenizer().encode(text)
+        tokens = [x + self.num_image_tokens for x in tmp]
+        return tokens if add_dummy_prefix else tokens[2:]
+
+    def postprocess(self, text):
+        text = text.replace("<n>", "\n")
+        text = text.replace(SPTokenizer.get_tab_token(), "\t")
+        for i in range(2, self.max_blank_length + 1):
+            text = text.replace(self.get_blank_token(i), " " * i)
+        return text
+
+    def decode(self, text_ids: List[int]) -> str:
+        ids = [int(_id) - self.num_image_tokens for _id in text_ids]
+        ids = [_id for _id in ids if _id >= 0]
+        text = self._get_text_tokenizer().decode(ids)
+        text = self.postprocess(text)
+        return text
+
+    def decode_tokens(self, tokens: List[str]) -> str:
+        text = self._get_text_tokenizer().convert_tokens_to_string(tokens)
+        text = self.postprocess(text)
+        return text
+
+    def tokenize(self, text: str, linebreak=True, whitespaces=True, add_dummy_prefix=True) -> List[str]:
+        """
+        @param text: Text to encode.
+        @param linebreak: Whether to encode newline (\n) in text.
+        @param whitespaces: Whether to encode multiple whitespaces or tab in text, useful for source code encoding.
+        @param special_tokens: Whether to encode special token ([MASK], [gMASK], etc.) in text.
+        @param add_dummy_prefix: Whether to add dummy blank space in the beginning.
+        """
+        text = self._preprocess(text, linebreak, whitespaces)
+        if not add_dummy_prefix:
+            text = "<n>" + text
+        tokens = self._get_text_tokenizer().tokenize(text)
+        return tokens if add_dummy_prefix else tokens[2:]
+
+    def __getitem__(self, x: Union[int, str]):
+        if isinstance(x, int):
+            if x < self.num_image_tokens:
+                return "<image_{}>".format(x)
+            else:
+                return self.text_tokenizer.convert_id_to_token(x - self.num_image_tokens)
+        elif isinstance(x, str):
+            if x.startswith("<image_") and x.endswith(">") and x[7:-1].isdigit():
+                return int(x[7:-1])
+            else:
+                return self.text_tokenizer.convert_token_to_id(x) + self.num_image_tokens
+        else:
+            raise ValueError("The key should be str or int.")
+
+
+class ChatGLMTokenizer(PreTrainedTokenizer):
+    """
+    Construct a ChatGLM tokenizer. Based on byte-level Byte-Pair-Encoding.
+
+    Args:
+        vocab_file (`str`):
+            Path to the vocabulary file.
+    """
+
+    vocab_files_names = {"vocab_file": "ice_text.model"}
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    model_input_names = ["input_ids", "attention_mask", "position_ids"]
+
+    def __init__(
+        self,
+        vocab_file,
+        do_lower_case=False,
+        remove_space=False,
+        bos_token="<sop>",
+        eos_token="<eop>",
+        end_token="</s>",
+        mask_token="[MASK]",
+        gmask_token="[gMASK]",
+        padding_side="left",
+        pad_token="<pad>",
+        unk_token="<unk>",
+        num_image_tokens=20000,
+        **kwargs,
+    ) -> None:
+        super().__init__(
+            do_lower_case=do_lower_case,
+            remove_space=remove_space,
+            padding_side=padding_side,
+            bos_token=bos_token,
+            eos_token=eos_token,
+            end_token=end_token,
+            mask_token=mask_token,
+            gmask_token=gmask_token,
+            pad_token=pad_token,
+            unk_token=unk_token,
+            num_image_tokens=num_image_tokens,
+            **kwargs,
+        )
+
+        self.do_lower_case = do_lower_case
+        self.remove_space = remove_space
+        self.vocab_file = vocab_file
+
+        self.bos_token = bos_token
+        self.eos_token = eos_token
+        self.end_token = end_token
+        self.mask_token = mask_token
+        self.gmask_token = gmask_token
+
+        self.sp_tokenizer = SPTokenizer(vocab_file, num_image_tokens=num_image_tokens)
+
+        """ Initialisation """
+
+    @property
+    def gmask_token_id(self) -> Optional[int]:
+        if self.gmask_token is None:
+            return None
+        return self.convert_tokens_to_ids(self.gmask_token)
+
+    @property
+    def end_token_id(self) -> Optional[int]:
+        """
+        `Optional[int]`: Id of the end of context token in the vocabulary. Returns `None` if the token has not been
+        set.
+        """
+        if self.end_token is None:
+            return None
+        return self.convert_tokens_to_ids(self.end_token)
+
+    @property
+    def vocab_size(self):
+        """Returns vocab size"""
+        return self.sp_tokenizer.num_tokens
+
+    def get_vocab(self):
+        """Returns vocab as a dict"""
+        vocab = {self._convert_id_to_token(i): i for i in range(self.vocab_size)}
+        vocab.update(self.added_tokens_encoder)
+        return vocab
+
+    def preprocess_text(self, inputs):
+        if self.remove_space:
+            outputs = " ".join(inputs.strip().split())
+        else:
+            outputs = inputs
+
+        if self.do_lower_case:
+            outputs = outputs.lower()
+
+        return outputs
+
+    def _tokenize(self, text, **kwargs):
+        """Returns a tokenized string."""
+        text = self.preprocess_text(text)
+
+        seq = self.sp_tokenizer.tokenize(text)
+
+        return seq
+
+    def convert_tokens_to_string(self, tokens: List[str]) -> str:
+        return self.sp_tokenizer.decode_tokens(tokens)
+
+    def _decode(self, token_ids: Union[int, List[int]], **kwargs) -> str:
+        if isinstance(token_ids, int):
+            token_ids = [token_ids]
+        if len(token_ids) == 0:
+            return ""
+        if self.pad_token_id in token_ids:  # remove pad
+            token_ids = list(filter((self.pad_token_id).__ne__, token_ids))
+        return super()._decode(token_ids, **kwargs)
+
+    def _convert_token_to_id(self, token):
+        """Converts a token (str) in an id using the vocab."""
+        return self.sp_tokenizer[token]
+
+    def _convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (str) using the vocab."""
+        return self.sp_tokenizer[index]
+
+    def save_vocabulary(self, save_directory, filename_prefix=None):
+        """
+        Save the vocabulary and special tokens file to a directory.
+
+        Args:
+            save_directory (`str`):
+                The directory in which to save the vocabulary.
+            filename_prefix (`str`, *optional*):
+                An optional prefix to add to the named of the saved files.
+
+        Returns:
+            `Tuple(str)`: Paths to the files saved.
+        """
+        if os.path.isdir(save_directory):
+            vocab_file = os.path.join(save_directory, self.vocab_files_names["vocab_file"])
+        else:
+            vocab_file = save_directory
+
+        with open(self.vocab_file, "rb") as fin:
+            proto_str = fin.read()
+
+        with open(vocab_file, "wb") as writer:
+            writer.write(proto_str)
+
+        return (vocab_file,)
+
+    def build_inputs_with_special_tokens(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
+        """
+        Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
+        adding special tokens. A BERT sequence has the following format:
+
+        - single sequence: `[CLS] X [SEP]`
+        - pair of sequences: `[CLS] A [SEP] B [SEP]`
+
+        Args:
+            token_ids_0 (`List[int]`):
+                List of IDs to which the special tokens will be added.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            `List[int]`: List of [input IDs](../glossary#input-ids) with the appropriate special tokens.
+        """
+        gmask_id = self.sp_tokenizer[self.gmask_token]
+        self.sp_tokenizer[self.eos_token]
+        token_ids_0 = token_ids_0 + [gmask_id, self.sp_tokenizer[self.bos_token]]
+        if token_ids_1 is not None:
+            token_ids_0 = token_ids_0 + token_ids_1
+        return token_ids_0
+
+    def _pad(
+        self,
+        encoded_inputs: Union[Dict[str, EncodedInput], BatchEncoding],
+        max_length: Optional[int] = None,
+        padding_strategy: PaddingStrategy = PaddingStrategy.DO_NOT_PAD,
+        pad_to_multiple_of: Optional[int] = None,
+        return_attention_mask: Optional[bool] = None,
+    ) -> dict:
+        """
+        Pad encoded inputs (on left/right and up to predefined length or max length in the batch)
+
+        Args:
+            encoded_inputs:
+                Dictionary of tokenized inputs (`List[int]`) or batch of tokenized inputs (`List[List[int]]`).
+            max_length: maximum length of the returned list and optionally padding length (see below).
+                Will truncate by taking into account the special tokens.
+            padding_strategy: PaddingStrategy to use for padding.
+
+                - PaddingStrategy.LONGEST Pad to the longest sequence in the batch
+                - PaddingStrategy.MAX_LENGTH: Pad to the max length (default)
+                - PaddingStrategy.DO_NOT_PAD: Do not pad
+                The tokenizer padding sides are defined in self.padding_side:
+
+                    - 'left': pads on the left of the sequences
+                    - 'right': pads on the right of the sequences
+            pad_to_multiple_of: (optional) Integer if set will pad the sequence to a multiple of the provided value.
+                This is especially useful to enable the use of Tensor Core on NVIDIA hardware with compute capability
+                `>= 7.5` (Volta).
+            return_attention_mask:
+                (optional) Set to False to avoid returning attention mask (default: set to model specifics)
+        """
+        # Load from model defaults
+        bos_token_id = self.sp_tokenizer[self.bos_token]
+        mask_token_id = self.sp_tokenizer[self.mask_token]
+        gmask_token_id = self.sp_tokenizer[self.gmask_token]
+        assert self.padding_side == "left"
+
+        required_input = encoded_inputs[self.model_input_names[0]]
+        seq_length = len(required_input)
+
+        if padding_strategy == PaddingStrategy.LONGEST:
+            max_length = len(required_input)
+
+        if max_length is not None and pad_to_multiple_of is not None and (max_length % pad_to_multiple_of != 0):
+            max_length = ((max_length // pad_to_multiple_of) + 1) * pad_to_multiple_of
+
+        needs_to_be_padded = padding_strategy != PaddingStrategy.DO_NOT_PAD and len(required_input) != max_length
+
+        # Initialize attention mask if not present.
+        if max_length is not None:
+            if "attention_mask" not in encoded_inputs:
+                if bos_token_id in required_input:
+                    context_length = required_input.index(bos_token_id)
+                else:
+                    context_length = seq_length
+                attention_mask = np.ones((1, seq_length, seq_length))
+                attention_mask = np.tril(attention_mask)
+                attention_mask[:, :, :context_length] = 1
+                attention_mask = np.bool_(attention_mask < 0.5)
+                encoded_inputs["attention_mask"] = attention_mask
+
+            if "position_ids" not in encoded_inputs:
+                if bos_token_id in required_input:
+                    context_length = required_input.index(bos_token_id)
+                else:
+                    context_length = seq_length
+                position_ids = np.arange(seq_length, dtype=np.int64)
+                mask_token = mask_token_id if mask_token_id in required_input else gmask_token_id
+                if mask_token in required_input:
+                    mask_position = required_input.index(mask_token)
+                    position_ids[context_length:] = mask_position
+                block_position_ids = np.concatenate(
+                    [
+                        np.zeros(context_length, dtype=np.int64),
+                        np.arange(1, seq_length - context_length + 1, dtype=np.int64),
+                    ]
+                )
+                encoded_inputs["position_ids"] = np.stack([position_ids, block_position_ids], axis=0)
+
+        if needs_to_be_padded:
+            difference = max_length - len(required_input)
+
+            if "attention_mask" in encoded_inputs:
+                encoded_inputs["attention_mask"] = np.pad(
+                    encoded_inputs["attention_mask"],
+                    pad_width=[(0, 0), (difference, 0), (difference, 0)],
+                    mode="constant",
+                    constant_values=True,
+                )
+            if "token_type_ids" in encoded_inputs:
+                encoded_inputs["token_type_ids"] = [self.pad_token_type_id] * difference + encoded_inputs[
+                    "token_type_ids"
+                ]
+            if "special_tokens_mask" in encoded_inputs:
+                encoded_inputs["special_tokens_mask"] = [1] * difference + encoded_inputs["special_tokens_mask"]
+            if "position_ids" in encoded_inputs:
+                encoded_inputs["position_ids"] = np.pad(
+                    encoded_inputs["position_ids"], pad_width=[(0, 0), (difference, 0)]
+                )
+            encoded_inputs[self.model_input_names[0]] = [self.pad_token_id] * difference + required_input
+
+        return encoded_inputs

applications/Chat/coati/models/chatglm/configuration_chatglm.py

+"""
+This code is copied from https://huggingface.co/THUDM/chatglm-6b/resolve/main/configuration_chatglm.py
+"""
+
+""" ChatGLM model configuration """
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+
+class ChatGLMConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`~ChatGLMModel`].
+    It is used to instantiate an ChatGLM model according to the specified arguments, defining the model
+    architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of
+    the ChatGLM-6B [THUDM/ChatGLM-6B](https://huggingface.co/THUDM/chatglm-6b) architecture.
+
+    Configuration objects inherit from  [`PretrainedConfig`] and can be used
+    to control the model outputs. Read the documentation from  [`PretrainedConfig`]
+    for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 150528):
+            Vocabulary size of the ChatGLM-6B model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`~ChatGLMModel`] or
+            [`~TFChatGLMModel`].
+        hidden_size (`int`, *optional*, defaults to 4096):
+            Dimension of the encoder layers and the pooler layer.
+        num_hidden_layers (`int`, *optional*, defaults to 28):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        inner_hidden_size (`int`, *optional*, defaults to 16384):
+            Dimension of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
+        max_sequence_length (`int`, *optional*, defaults to 512):
+            The maximum sequence length that this model might ever be used with.
+            Typically set this to something large just in case (e.g., 512 or 1024 or 2048).
+        layernorm_epsilon (`float`, *optional*, defaults to 1e-5):
+            The epsilon used by the layer normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether the model should return the last key/values attentions (not used by all models).
+        Example:
+
+    ```python
+    >>> from configuration_chatglm import ChatGLMConfig
+    >>> from modeling_chatglm import ChatGLMModel
+
+    >>> # Initializing a ChatGLM-6B THUDM/ChatGLM-6B style configuration
+    >>> configuration = ChatGLMConfig()
+
+    >>> # Initializing a model from the THUDM/ChatGLM-6B style configuration
+    >>> model = ChatGLMModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+    model_type = "chatglm"
+
+    def __init__(
+        self,
+        vocab_size=130528,
+        hidden_size=4096,
+        num_layers=28,
+        num_attention_heads=32,
+        layernorm_epsilon=1e-5,
+        use_cache=True,
+        bos_token_id=130004,
+        eos_token_id=130005,
+        mask_token_id=130000,
+        gmask_token_id=130001,
+        pad_token_id=3,
+        max_sequence_length=2048,
+        inner_hidden_size=16384,
+        position_encoding_2d=True,
+        quantization_bit=0,
+        pre_seq_len=None,
+        prefix_projection=False,
+        **kwargs,
+    ):
+        self.num_layers = num_layers
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.num_attention_heads = num_attention_heads
+        self.max_sequence_length = max_sequence_length
+        self.layernorm_epsilon = layernorm_epsilon
+        self.inner_hidden_size = inner_hidden_size
+        self.use_cache = use_cache
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+        self.pad_token_id = pad_token_id
+        self.mask_token_id = mask_token_id
+        self.gmask_token_id = gmask_token_id
+        self.position_encoding_2d = position_encoding_2d
+        self.quantization_bit = quantization_bit
+        self.pre_seq_len = pre_seq_len
+        self.prefix_projection = prefix_projection
+
+        super().__init__(pad_token_id=pad_token_id, bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)

applications/Chat/coati/models/deberta/__init__.py

-from .deberta_critic import DebertaCritic
-from .deberta_rm import DebertaRM
-
-__all__ = ['DebertaCritic', 'DebertaRM']

applications/Chat/coati/models/deberta/deberta_critic.py

-from typing import Optional
-
-import torch.nn as nn
-from transformers import DebertaV2Config, DebertaV2Model
-
-from ..base import Critic
-
-
-class DebertaCritic(Critic):
-    """
-    Deberta Critic model.
-
-    Args:
-        pretrained (str): Pretrained model name or path.
-        config (DebertaV2Config): Model config.
-        checkpoint (bool): Enable gradient checkpointing.
-        lora_rank (int): Rank of the LO-RA decomposition.
-        lora_train_bias (str): LoRA bias training mode.
-    """
-
-    def __init__(self,
-                 pretrained: Optional[str] = None,
-                 config: Optional[DebertaV2Config] = None,
-                 checkpoint: bool = False,
-                 lora_rank: int = 0,
-                 lora_train_bias: str = 'none') -> None:
-        if pretrained is not None:
-            model = DebertaV2Model.from_pretrained(pretrained)
-        elif config is not None:
-            model = DebertaV2Model(config)
-        else:
-            model = DebertaV2Model(DebertaV2Config())
-        if checkpoint:
-            model.gradient_checkpointing_enable()
-        value_head = nn.Linear(model.config.hidden_size, 1)
-        super().__init__(model, value_head, lora_rank, lora_train_bias)

applications/Chat/coati/models/deberta/deberta_rm.py

-from typing import Optional
-
-import torch.nn as nn
-from transformers import DebertaV2Config, DebertaV2Model
-
-from ..base import RewardModel
-
-
-class DebertaRM(RewardModel):
-    """
-    Deberta Reward model.
-
-    Args:
-        pretrained (str): Pretrained model name or path.
-        config (DebertaV2Config): Model config.
-        checkpoint (bool): Enable gradient checkpointing.
-        lora_rank (int): Rank of the LO-RA decomposition.
-        lora_train_bias (str): LoRA bias training mode.
-    """
-
-    def __init__(self,
-                 pretrained: str = None,
-                 config: Optional[DebertaV2Config] = None,
-                 checkpoint: bool = False,
-                 lora_rank: int = 0,
-                 lora_train_bias: str = 'none') -> None:
-        if pretrained is not None:
-            model = DebertaV2Model.from_pretrained(pretrained)
-        elif config is not None:
-            model = DebertaV2Model(config)
-        else:
-            model = DebertaV2Model(DebertaV2Config())
-        if checkpoint:
-            model.gradient_checkpointing_enable()
-        value_head = nn.Linear(model.config.hidden_size, 1)
-        value_head.weight.data.normal_(mean=0.0, std=1 / (model.config.hidden_size + 1))
-        super().__init__(model, value_head, lora_rank, lora_train_bias)

applications/Chat/coati/models/generation.py

@@ -2,7 +2,9 @@ from typing import Any, Callable, Optional
 
 import torch
 import torch.distributed as dist
-import torch.nn as nn
+from transformers import PreTrainedTokenizer
+
+from .base import Actor
 
 try:
     from transformers.generation_logits_process import (
@@ -15,9 +17,9 @@ except ImportError:
     from transformers.generation import LogitsProcessorList, TemperatureLogitsWarper, TopKLogitsWarper, TopPLogitsWarper
 
 
-def prepare_logits_processor(top_k: Optional[int] = None,
-                             top_p: Optional[float] = None,
-                             temperature: Optional[float] = None) -> LogitsProcessorList:
+def _prepare_logits_processor(
+    top_k: Optional[int] = None, top_p: Optional[float] = None, temperature: Optional[float] = None
+) -> LogitsProcessorList:
     processor_list = LogitsProcessorList()
     if temperature is not None and temperature != 1.0:
         processor_list.append(TemperatureLogitsWarper(temperature))
@@ -36,7 +38,8 @@ def _is_sequence_finished(unfinished_sequences: torch.Tensor) -> bool:
     return unfinished_sequences.max() == 0
 
 
-def sample(model: nn.Module,
+def _sample(
+    model: Actor,
     input_ids: torch.Tensor,
     max_length: int,
     early_stopping: bool = False,
@@ -47,21 +50,22 @@ def sample(model: nn.Module,
     temperature: Optional[float] = None,
     prepare_inputs_fn: Optional[Callable[[torch.Tensor, Any], dict]] = None,
     update_model_kwargs_fn: Optional[Callable[[dict, Any], dict]] = None,
-           **model_kwargs) -> torch.Tensor:
+    **model_kwargs,
+) -> torch.Tensor:
     if input_ids.size(1) >= max_length:
         return input_ids
 
-    logits_processor = prepare_logits_processor(top_k, top_p, temperature)
+    logits_processor = _prepare_logits_processor(top_k, top_p, temperature)
     unfinished_sequences = input_ids.new(input_ids.shape[0]).fill_(1)
 
     for _ in range(input_ids.size(1), max_length):
-        model_inputs = prepare_inputs_fn(input_ids, **model_kwargs) if prepare_inputs_fn is not None else {
-            'input_ids': input_ids
-        }
+        model_inputs = (
+            prepare_inputs_fn(input_ids, **model_kwargs) if prepare_inputs_fn is not None else {"input_ids": input_ids}
+        )
         outputs = model(**model_inputs)
 
-        next_token_logits = outputs['logits'][:, -1, :]
-        # pre-process distribution
+        # NOTE: this is correct only in left padding mode
+        next_token_logits = outputs["logits"][:, -1, :]
         next_token_logits = logits_processor(input_ids, next_token_logits)
         # sample
         probs = torch.softmax(next_token_logits, dim=-1, dtype=torch.float)
@@ -69,8 +73,7 @@ def sample(model: nn.Module,
 
         # finished sentences should have their next token be a padding token
         if eos_token_id is not None:
-            if pad_token_id is None:
-                raise ValueError("If `eos_token_id` is defined, make sure that `pad_token_id` is defined.")
+            assert pad_token_id is not None, "If `eos_token_id` is defined, make sure that `pad_token_id` is defined."
             next_tokens = next_tokens * unfinished_sequences + pad_token_id * (1 - unfinished_sequences)
 
         # update generated ids, model inputs for next step
@@ -89,20 +92,22 @@ def sample(model: nn.Module,
     return input_ids
 
 
-def generate(model: nn.Module,
+@torch.no_grad()
+def generate(
+    model: Actor,
     input_ids: torch.Tensor,
+    tokenizer: PreTrainedTokenizer,
     max_length: int,
     num_beams: int = 1,
     do_sample: bool = True,
     early_stopping: bool = False,
-             eos_token_id: Optional[int] = None,
-             pad_token_id: Optional[int] = None,
     top_k: Optional[int] = None,
     top_p: Optional[float] = None,
     temperature: Optional[float] = None,
     prepare_inputs_fn: Optional[Callable[[torch.Tensor, Any], dict]] = None,
     update_model_kwargs_fn: Optional[Callable[[dict, Any], dict]] = None,
-             **model_kwargs) -> torch.Tensor:
+    **model_kwargs,
+) -> torch.Tensor:
     """Generate token sequence. The returned sequence is input_ids + generated_tokens.
 
     Args:
@@ -112,34 +117,35 @@ def generate(model: nn.Module,
         num_beams (int, optional): number of beams. Defaults to 1.
         do_sample (bool, optional): whether to do sample. Defaults to True.
         early_stopping (bool, optional): if True, the sequence length may be smaller than max_length due to finding eos. Defaults to False.
-        eos_token_id (Optional[int], optional): end of sequence token id. Defaults to None.
-        pad_token_id (Optional[int], optional): pad token id. Defaults to None.
         top_k (Optional[int], optional): the number of highest probability vocabulary tokens to keep for top-k-filtering. Defaults to None.
         top_p (Optional[float], optional): If set to float < 1, only the smallest set of most probable tokens with probabilities that add up to top_p or higher are kept for generation. Defaults to None.
         temperature (Optional[float], optional): The value used to module the next token probabilities. Defaults to None.
         prepare_inputs_fn (Optional[Callable[[torch.Tensor, Any], dict]], optional): Function to preprocess model inputs. Arguments of this function should be input_ids and model_kwargs. Defaults to None.
         update_model_kwargs_fn (Optional[Callable[[dict, Any], dict]], optional): Function to update model_kwargs based on outputs. Arguments of this function should be outputs and model_kwargs. Defaults to None.
     """
-    is_greedy_gen_mode = ((num_beams == 1) and do_sample is False)
-    is_sample_gen_mode = ((num_beams == 1) and do_sample is True)
-    is_beam_gen_mode = ((num_beams > 1) and do_sample is False)
+    assert tokenizer.padding_side == "left", "Current generation only supports left padding."
+    is_greedy_gen_mode = (num_beams == 1) and do_sample is False
+    is_sample_gen_mode = (num_beams == 1) and do_sample is True
+    is_beam_gen_mode = (num_beams > 1) and do_sample is False
     if is_greedy_gen_mode:
         # run greedy search
         raise NotImplementedError
     elif is_sample_gen_mode:
         # run sample
-        return sample(model,
+        return _sample(
+            model,
             input_ids,
             max_length,
             early_stopping=early_stopping,
-                      eos_token_id=eos_token_id,
-                      pad_token_id=pad_token_id,
+            eos_token_id=tokenizer.eos_token_id,
+            pad_token_id=tokenizer.pad_token_id,
             top_k=top_k,
             top_p=top_p,
             temperature=temperature,
             prepare_inputs_fn=prepare_inputs_fn,
             update_model_kwargs_fn=update_model_kwargs_fn,
-                      **model_kwargs)
+            **model_kwargs,
+        )
     elif is_beam_gen_mode:
         raise NotImplementedError
     else:

applications/Chat/coati/models/gpt/__init__.py

@@ -2,4 +2,4 @@ from .gpt_actor import GPTActor
 from .gpt_critic import GPTCritic
 from .gpt_rm import GPTRM
 
-__all__ = ['GPTActor', 'GPTCritic', 'GPTRM']
+__all__ = ["GPTActor", "GPTCritic", "GPTRM"]

applications/Chat/coati/models/gpt/gpt_actor.py

@@ -18,13 +18,15 @@ class GPTActor(Actor):
         lora_train_bias (str): Bias training strategy for the LoRa layer.
     """
 
-    def __init__(self,
+    def __init__(
+        self,
         pretrained: Optional[str] = None,
         config: Optional[GPT2Config] = None,
         checkpoint: bool = False,
         lora_rank: int = 0,
-                 lora_train_bias: str = 'none',
-                 **kwargs) -> None:
+        lora_train_bias: str = "none",
+        **kwargs,
+    ) -> None:
         if pretrained is not None:
             model = GPT2LMHeadModel.from_pretrained(pretrained)
         elif config is not None:

applications/Chat/coati/models/llama/__init__.py

@@ -2,4 +2,4 @@ from .llama_actor import LlamaActor
 from .llama_critic import LlamaCritic
 from .llama_rm import LlamaRM
 
-__all__ = ['LlamaActor', 'LlamaCritic', 'LlamaRM']
+__all__ = ["LlamaActor", "LlamaCritic", "LlamaRM"]