microsoft

/

Phi-3-mini-128k-instruct

like 1.57k

Hi, I'm doing instruction tuning of Phi-3-mini-128k-instruct within the LLaVA-1.5 framework. However, I'm facing Tokenization Mismatch errors. I'm using the following preprocess function:

def preprocess_phi3(
    sources,
    tokenizer: transformers.PreTrainedTokenizer,
    has_image: bool = False
) -> Dict:
    conv = conversation_lib.default_conversation.copy()
    roles = {"human": conv.roles[0], "gpt": conv.roles[1]}

    # Apply prompt templates
    conversations = []
    for i, source in enumerate(sources):
        if roles[source[0]["from"]] != conv.roles[0]:
            # Skip the first one if it is not from human
            source = source[1:]

        conv.messages = []
        for j, sentence in enumerate(source):
            role = roles[sentence["from"]]
            assert role == conv.roles[j % 2], f"{i}"
            conv.append_message(role, sentence["value"])
        conversations.append(conv.get_prompt())

    # Tokenize conversations

    if has_image:
        input_ids = torch.stack([tokenizer_image_token(prompt, tokenizer, return_tensors='pt') for prompt in conversations], dim=0)
    else:
        input_ids = tokenizer(
            conversations,
            return_tensors="pt",
            padding="longest",
            max_length=tokenizer.model_max_length,
            truncation=True,
        ).input_ids

    targets = input_ids.clone()
    assert conv.sep_style == conversation_lib.SeparatorStyle.MPT

    # Mask targets
    sep = conv.sep + conv.roles[1]
    for conversation, target in zip(conversations, targets):
        total_len = int(target.ne(tokenizer.pad_token_id).sum())

        rounds = conversation.split(conv.sep)
        re_rounds = [conv.sep.join(rounds[:3])]  # system + user + gpt
        for conv_idx in range(3, len(rounds), 2):
            re_rounds.append(conv.sep.join(rounds[conv_idx:conv_idx+2]))    # user + gpt
        cur_len = 0
        target[:cur_len] = IGNORE_INDEX
        for i, rou in enumerate(re_rounds):
            if rou == "":
                break

            parts = rou.split(sep)
            if len(parts) != 2:
                break
            parts[0] += sep

            if has_image:
                round_len = len(tokenizer_image_token(rou, tokenizer))
                instruction_len = len(tokenizer_image_token(parts[0], tokenizer)) - 1
            else:
                round_len = len(tokenizer(rou).input_ids)
                instruction_len = len(tokenizer(parts[0]).input_ids) - 1

            if i == 0:
                round_len += 1
                instruction_len += 1
            else:
                round_len -= 2
                instruction_len -= 2

            if i != 0 and getattr(tokenizer, 'legacy', False) and IS_TOKENIZER_GREATER_THAN_0_14:
                round_len += 1
                instruction_len += 1

            target[cur_len : cur_len + instruction_len] = IGNORE_INDEX

            cur_len += round_len
        target[cur_len:] = IGNORE_INDEX

        if cur_len < tokenizer.model_max_length:
            if cur_len != total_len:
                target[:] = IGNORE_INDEX
                print(
                    f"WARNING: tokenization mismatch: {cur_len} vs. {total_len}."
                    f" (ignored)"
                )

    return dict(
        input_ids=input_ids,
        labels=targets,
    )

It would be great to get some insights into this!