hf_causal.py

import torch
import transformers

import copy
from tqdm import tqdm

import torch.nn.functional as F

from lm_eval import utils
from lm_eval.logger import eval_logger
from lm_eval.api.model import LM
from lm_eval.api.registry import register_model

from lm_eval.utils import MultiTokenEOSCriteria, stop_sequences_criteria

from accelerate import Accelerator
from typing import Optional, Union


@register_model("hf-causal")
class HFCausalLM(LM):
    def __init__(
        self,
        device="cuda",
        pretrained="gpt2",
        revision="main",
        low_cpu_mem_usage=None,
        dtype: Optional[Union[str, torch.dtype]] = "auto",
        subfolder=None,
        tokenizer=None,
        batch_size=1,
    ):
        super().__init__()

        assert isinstance(device, str)
        assert isinstance(pretrained, str)
        assert isinstance(batch_size, int)

        gpus = torch.cuda.device_count()

        if gpus <= 1:
            if device:
                if device not in ["cuda", "cpu"]:
                    device = int(device)
                self._device = torch.device(device)
                eval_logger.info(f"Using device '{device}'")
            else:
                eval_logger.info("Device not specified")
                eval_logger.info(f"Cuda Available? {torch.cuda.is_available()}")
                self._device = (
                    torch.device("cuda")
                    if torch.cuda.is_available()
                    else torch.device("cpu")
                )
            self._rank = 0
            self._world_size = 1

        else:
            self._device = "cpu"

        # TODO: update this to be less of a hack once subfolder is fixed in HF
        revision = revision + ("/" + subfolder if subfolder is not None else "")

        self.model = transformers.AutoModelForCausalLM.from_pretrained(
            pretrained,
            revision=revision,
            low_cpu_mem_usage=low_cpu_mem_usage,
            torch_dtype=utils.get_dtype(dtype),
        ).to(self.device)
        self.model.eval()

        eval_logger.info(self.model.dtype)

        self.tokenizer = transformers.AutoTokenizer.from_pretrained(
            pretrained if tokenizer is None else tokenizer,
            revision=revision,
        )

        self.vocab_size = self.tokenizer.vocab_size

        # multithreading and batching
        self.batch_size_per_gpu = batch_size  # todo: adaptive batch size

        # multigpu support with accelerate
        if gpus > 1:
            accelerator = Accelerator()
            if gpus > accelerator.num_processes:
                eval_logger.warning(
                    "WARNING: The number of total system GPUs does not match the number of spawned processes. "
                    "If you would like to use data parallelism, please launch the script "
                    "with 'accelerate launch *script*'. "
                    f"Current run will proceed with {accelerator.num_processes} devices."
                )
                self._rank = accelerator.local_process_index
                self._world_size = accelerator.num_processes
                # manually set model to use gpu, for case where many GPUs available but
                # only seek to use one
                self._device = (
                    torch.device(f"cuda:{accelerator.local_process_index}")
                    if torch.cuda.is_available()
                    else torch.device("cpu")
                )
                self.model.to(self.device)
            else:
                self.model = accelerator.prepare(self.model)
                self._device = torch.device(f"cuda:{accelerator.local_process_index}")
                self.accelerator = accelerator

                if self.accelerator.is_local_main_process:
                    eval_logger.info(f"Using {gpus} devices with data parallelism")

                self._rank = self.accelerator.local_process_index
                self._world_size = self.accelerator.num_processes

    @property
    def eot_token_id(self):
        # we use EOT because end of *text* is more accurate for what we're doing than end of *sentence*
        return self.tokenizer.eos_token_id

    @property
    def max_length(self):
        try:
            if hasattr(self, "accelerator"):
                return self.accelerator.unwrap_model(self.model).config.n_ctx
            else:
                return self.model.config.n_ctx
        except AttributeError:
            # gptneoconfig doesn't have n_ctx apparently
            if hasattr(self, "accelerator"):
                return self.accelerator.unwrap_model(
                    self.model
                ).config.max_position_embeddings
            else:
                return self.model.config.max_position_embeddings

    @property
    def max_gen_toks(self):
        return 256

    @property
    def batch_size(self):
        return self.batch_size_per_gpu

    @property
    def device(self):
        return self._device

    @property
    def rank(self):
        return self._rank

    @property
    def world_size(self):
        return self._world_size

    def tok_encode(self, string: str):
        return self.tokenizer.encode(string, add_special_tokens=False)

    def tok_decode(self, tokens):
        return self.tokenizer.decode(tokens)

    def _model_call(self, inps):
        """
        inps: a torch tensor of shape [batch, sequence]
        the size of sequence may vary from call to call

        returns: a torch tensor of shape [batch, sequence, vocab] with the
        logits returned from the model
        """
        with torch.no_grad():
            return self.model(inps).logits

    def _model_generate(self, context, max_length, stop, **generation_kwargs):
        # we require users to pass do_sample=True explicitly
        # for non-greedy gen. This should be reevaluated when considering beam search.
        if "do_sample" not in generation_kwargs.keys():
            generation_kwargs["do_sample"] = False
        # build stopping criteria
        stopping_criteria = stop_sequences_criteria(
            self.tokenizer, stop, 1, context.shape[0]
        )
        if hasattr(self, "accelerator"):
            return self.accelerator.unwrap_model(self.model).generate(
                context,
                max_length=max_length,
                stopping_criteria=stopping_criteria,
                pad_token_id=self.eot_token_id,
                use_cache=True,
                **generation_kwargs,
            )
        else:
            return self.model.generate(
                context,
                max_length=max_length,
                stopping_criteria=stopping_criteria,
                pad_token_id=self.eot_token_id,
                use_cache=True,
                **generation_kwargs,
            )

    def loglikelihood(self, requests):
        new_reqs = []
        for context, continuation in [req.args for req in requests]:
            if context == "":
                # end of text as context
                context_enc = [self.eot_token_id]
            else:
                context_enc = self.tok_encode(context)

            continuation_enc = self.tok_encode(continuation)

            new_reqs.append(((context, continuation), context_enc, continuation_enc))

        return self._loglikelihood_tokens(new_reqs)

    def loglikelihood_rolling(self, requests):
        loglikelihoods = []
        for (string,) in tqdm([req.args for req in requests], disable=(self.rank != 0)):
            rolling_token_windows = list(
                map(
                    utils.make_disjoint_window,
                    utils.get_rolling_token_windows(
                        token_list=self.tok_encode(string),
                        prefix_token=self.eot_token_id,
                        max_seq_len=self.max_length,
                        context_len=1,
                    ),
                )
            )

            rolling_token_windows = [(None,) + x for x in rolling_token_windows]

            # TODO: extract out this call so it only gets called once and also somehow figure out partial caching for
            # that

            pad_amnt = 0
            if self.world_size > 1:
                # TODO: Comment on what we do here
                mytensor = torch.tensor(len(rolling_token_windows), device=self.device)
                gathered = (
                    self.accelerator.gather(mytensor).cpu().detach().numpy().tolist()
                )

                pad_amnt = max(gathered) - gathered[self.rank]
                if pad_amnt > 0:
                    rolling_token_windows += pad_amnt * [rolling_token_windows[0]]

            string_nll = self._loglikelihood_tokens(
                rolling_token_windows, disable_tqdm=True
            )

            if (self.world_size > 1) and (pad_amnt > 0):
                string_nll = [x[0] for x in string_nll[:-pad_amnt]]
            else:
                # discard is_greedy
                string_nll = [x[0] for x in string_nll]

            string_nll = sum(string_nll)
            loglikelihoods.append(string_nll)

        return loglikelihoods

    def _loglikelihood_tokens(self, requests, disable_tqdm=False):
        # TODO: implement some kind of efficient-request-middleware that lumps together requests with the same context
        res = []

        def _collate(x):
            # the negative sign on len(toks) sorts descending - this has a few advantages:
            # - time estimates will always be over not underestimates, which is more useful for planning
            # - to know the size of a batch when going through the list, you know the first one is always the batch
            #   padded context length. this is useful to simplify the batching logic and more importantly to make
            #   automatic adaptive batches much much easier to implement
            # - any OOMs will happen right away rather than near the end

            toks = x[1] + x[2]
            return -len(toks), tuple(toks)

        # TODO: automatic (variable) batch size detection for vectorization
        re_ord = utils.Reorderer(requests, _collate)
        for chunk in utils.chunks(
            tqdm(re_ord.get_reordered(), disable=(disable_tqdm or (self.rank != 0))),
            self.batch_size,
        ):

            inps = []
            cont_toks_list = []
            inplens = []

            padding_length = None

            # because vectorizing is annoying, we first convert each (context, continuation) pair to padded
            # tensors, then we pack them together into a batch, call the model, and then pick it all apart
            # again because vectorizing is annoying

            for _, context_enc, continuation_enc in chunk:
                # sanity check
                assert len(context_enc) > 0
                assert len(continuation_enc) > 0
                assert len(continuation_enc) <= self.max_length

                # how this all works:
                #          CTX      CONT
                # inp    0 1 2 3|4 5 6 7 8 9   <- last token is deleted by inp[:, :-1]
                # model  \               \
                # logits   1 2 3|4 5 6 7 8 9   <- the ctx half gets tossed out by the
                # cont_toks      4 5 6 7 8 9      [:, -len(continuation_enc):, :self.vocab_size] slice

                # when too long to fit in context, truncate from the left
                inp = torch.tensor(
                    (context_enc + continuation_enc)[-(self.max_length + 1) :][:-1],
                    dtype=torch.long,
                ).to(self.device)
                (inplen,) = inp.shape

                cont = continuation_enc

                # since in _collate we make sure length is descending, the longest is always the first one.
                padding_length = (
                    padding_length if padding_length is not None else inplen
                )

                # pad length from seq to padding_length
                inp = torch.cat(
                    [
                        inp,  # [seq]
                        torch.zeros(padding_length - inplen, dtype=torch.long).to(
                            inp.device
                        ),  # [padding_length - seq]
                    ],
                    dim=0,
                )

                inps.append(inp.unsqueeze(0))  # [1, padding_length]
                cont_toks_list.append(cont)
                inplens.append(inplen)

            batched_inps = torch.cat(inps, dim=0)  # [batch, padding_length
            multi_logits = F.log_softmax(
                self._model_call(batched_inps), dim=-1
            ).cpu()  # [batch, padding_length, vocab]

            for (cache_key, _, _), logits, inp, inplen, cont_toks in zip(
                chunk, multi_logits, inps, inplens, cont_toks_list
            ):

                # Slice to original seq length
                contlen = len(cont_toks)
                logits = logits[inplen - contlen : inplen].unsqueeze(
                    0
                )  # [1, seq, vocab]

                # Check if per-token argmax is exactly equal to continuation
                greedy_tokens = logits.argmax(dim=-1)
                cont_toks = torch.tensor(cont_toks, dtype=torch.long).unsqueeze(
                    0
                )  # [1, seq]
                max_equal = (greedy_tokens == cont_toks).all()

                # Obtain log-probs at the corresponding continuation token indices
                # last_token_slice = logits[:, -1, :].squeeze(0).tolist()
                logits = torch.gather(logits, 2, cont_toks.unsqueeze(-1)).squeeze(
                    -1
                )  # [1, seq]

                # Answer: (log prob, is-exact-match)
                answer = (float(logits.sum()), bool(max_equal))

                res.append(answer)

        return re_ord.get_original(res)

    def greedy_until(self, requests):
        # TODO: implement fully general `until` that handles until that are
        #       multiple tokens or that span multiple tokens correctly

        res = []

        def _collate(x):
            toks = self.tok_encode(x[0])
            return len(toks), x[0]

        re_ord = utils.Reorderer([req.args for req in requests], _collate)

        for context, gen_kwargs in tqdm(re_ord.get_reordered()):
            until = None
            if isinstance(gen_kwargs, dict):
                gen_kwargs = copy.deepcopy(gen_kwargs)  # edge case for repeats > 1
                if "until" in gen_kwargs.keys():
                    until = gen_kwargs.pop("until")
                    if isinstance(until, str):
                        until = [gen_kwargs]
                    elif not isinstance(until, list):
                        raise ValueError(
                            f"Expected `gen_kwargs['until']` to be of type Union[str,list] but got {until}"
                        )
            else:
                raise ValueError(
                    f"Expected `gen_kwargs` to be of type `dict` but got {gen_kwargs}"
                )
            if not until:
                until = [self.tok_decode(self.eot_token_id)]
            if "max_gen_toks" in gen_kwargs.keys():
                max_gen_toks = gen_kwargs.pop("max_gen_toks")
            else:
                max_gen_toks = self.max_gen_toks

            primary_until = until[0]
            # try:
            #     (primary_until,) = self.tok_encode(until[0])
            # except Exception:
            #     # if our primary until would be multiple tokens long, we'll have errors.
            #     # TODO: handling this better will let us stop generating earlier + often.
            #     primary_until = self.eot_token_id

            context_enc = torch.tensor(
                [self.tok_encode(context)[max_gen_toks - self.max_length :]]
            ).to(self.device)

            cont = self._model_generate(
                context=context_enc,
                max_length=context_enc.shape[1] + max_gen_toks,
                stop=primary_until,
                **gen_kwargs,
            )

            s = self.tok_decode(cont[0].tolist()[context_enc.shape[1] :])

            for term in until:
                s = s.split(term)[0]

            res.append(s)

        return re_ord.get_original(res)