Readme.md

Usage

Model loading

import torch
from torch import nn
from torch.nn import CrossEntropyLoss, MSELoss, BCEWithLogitsLoss
from transformers import LlamaPreTrainedModel,LlamaModel,Gemma2PreTrainedModel,Gemma2Model,Cache
from transformers.modeling_outputs import SequenceClassifierOutputWithPast
from typing import Optional, List, Union, Tuple

@dataclass
class Config:
    gemma_dir = '/kaggle/input/v7-dpo-16bit-01234-8bit-all/v7_dpo_16bit_01234_8bit_all'
    max_length = 2000
    batch_size = 8
    device = torch.device("cuda") if torch.cuda_is_available() else torch.device("cpu")

cfg = Config()

class Gemma2ForSequenceClassificationV1(Gemma2PreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels
        self.model = Gemma2Model(config)
        self.score = nn.Linear(config.hidden_size, self.num_labels, bias=False)

        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        return self.model.embed_tokens

    def set_input_embeddings(self, value):
        self.model.embed_tokens = value

    def forward(
            self,
            input_ids: torch.LongTensor = None,
            attention_mask: Optional[torch.Tensor] = None,
            position_ids: Optional[torch.LongTensor] = None,
            past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
            inputs_embeds: Optional[torch.FloatTensor] = None,
            labels: Optional[torch.LongTensor] = None,
            use_cache: Optional[bool] = None,
            output_attentions: Optional[bool] = None,
            output_hidden_states: Optional[bool] = None,
            return_dict: Optional[bool] = None,
    ) -> Union[Tuple, SequenceClassifierOutputWithPast]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        transformer_outputs = self.model(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )
        hidden_states = transformer_outputs[0]
#         logits = self.score(hidden_states)

        if input_ids is not None:
            batch_size = input_ids.shape[0]
        else:
            batch_size = inputs_embeds.shape[0]

        if self.config.pad_token_id is None and batch_size != 1:
            raise ValueError("Cannot handle batch sizes > 1 if no padding token is defined.")
        if self.config.pad_token_id is None:
            sequence_lengths = -1
        else:
            if input_ids is not None:
                # if no pad token found, use modulo instead of reverse indexing for ONNX compatibility
                sequence_lengths = torch.eq(input_ids, self.config.pad_token_id).int().argmax(-1) - 1
                sequence_lengths = sequence_lengths % input_ids.shape[-1]
                sequence_lengths = sequence_lengths.to(hidden_states.device)
            else:
                sequence_lengths = -1
        hidden_states = hidden_states[
            torch.arange(batch_size, device=hidden_states.device), sequence_lengths]  # eos
        pooled_logits = self.score(hidden_states)
        
        return pooled_logits


tokenizer = GemmaTokenizerFast.from_pretrained(cfg.gemma_dir)

model = Gemma2ForSequenceClassificationV1.from_pretrained(
    cfg.gemma_dir,
    num_labels=3,
    device_map=cfg.device,
    use_cache=False,
)
model.config.pad_token_id = tokenizer.pad_token_id

Inference

def create_rounds(query: str,
                  answer_a: str,
                  answer_b: str) -> str:
    prompt =f"""User question:
    \"""{query}\"""
    Answer A:
    \"""{answer_a}\"""
    Answer B:
    \"""{answer_b}\"""
    """
    return prompt


@torch.no_grad()
@torch.cuda.amp.autocast()
def single_prompt_inference(prompt, model, device, max_length=cfg.max_length):
    """
    Perform inference on a single prompt.

    Args:
        prompt (str): The input prompt for inference.
        model (torch.nn.Module): The model used for inference.
        device (torch.device): The device to run inference on.
        tokenizer (Tokenizer): Tokenizer for preprocessing input text.
        max_length (int): Maximum sequence length for tokenization.

    Returns:
        dict: Probabilities for "a_win", "b_win", and "tie".
    """
    # Tokenize the input prompt
    input_ids = tokenizer(prompt, truncation=True, max_length=max_length)['input_ids']
    input_ids.append(tokenizer.eos_token_id)

    # Prepare inputs
    inputs = pad_without_fast_tokenizer_warning(
        tokenizer,
        {"input_ids": [input_ids]},  # Wrap in a list for compatibility
        padding="max_length",
        pad_to_multiple_of=None,
        max_length=max_length,
        return_tensors="pt",
    )

    # Move inputs to the appropriate device
    inputs = inputs.to(cfg.device)

    # Run the model
    outputs = model(**inputs)

    # Get probabilities using softmax
    proba = outputs.softmax(-1).cpu().squeeze()

    return {
        "winner_model_a": proba[0].item(),
        "winner_model_b": proba[1].item(),
        "tie": proba[2].item(),
    }

query = "What is the height of the reassembled blind product?"
answer_a = "You can find all the technical information directly on the product sheet on our site."
answer_b = "The height of the aluminum Venetian blind is 130 cm."
prompt_direct = create_rounds(query, answer_a, answer_b)

single_prompt_inference(prompt_direct, model, device)

Credits to @sayoulala on kaggle for winnig the competition https://www.kaggle.com/competitions/lmsys-chatbot-arena and submitting this model.