hf_text-generation-inference/server/text_generation_server/models/model.py

115 lines
3.9 KiB
Python
Raw Normal View History

import inspect
import torch
from abc import ABC, abstractmethod
from typing import List, Tuple, Optional, TypeVar, Type
feat(server): Add exllama GPTQ CUDA kernel support #553 (#666) Just trying to get the integration tests to pass. # What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflects the extent of your awesome contribution. Then, please replace this with a description of the change and which issue is fixed (if applicable). Please also include relevant motivation and context. List any dependencies (if any) that are required for this change. Once you're done, someone will review your PR shortly (see the section "Who can review?" below to tag some potential reviewers). They may suggest changes to make the code even better. If no one reviewed your PR after a week has passed, don't hesitate to post a new comment @-mentioning the same persons---sometimes notifications get lost. --> <!-- Remove if not applicable --> Fixes # (issue) ## Before submitting - [ ] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). - [ ] Did you read the [contributor guideline](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md#start-contributing-pull-requests), Pull Request section? - [ ] Was this discussed/approved via a Github issue or the [forum](https://discuss.huggingface.co/)? Please add a link to it if that's the case. - [ ] Did you make sure to update the documentation with your changes? Here are the [documentation guidelines](https://github.com/huggingface/transformers/tree/main/docs), and [here are tips on formatting docstrings](https://github.com/huggingface/transformers/tree/main/docs#writing-source-documentation). - [ ] Did you write any new necessary tests? ## Who can review? Anyone in the community is free to review the PR once the tests have passed. Feel free to tag members/contributors who may be interested in your PR. <!-- Your PR will be replied to more quickly if you can figure out the right person to tag with @ @OlivierDehaene OR @Narsil --> --------- Co-authored-by: Felix Marty <9808326+fxmarty@users.noreply.github.com>
2023-07-21 02:59:00 -06:00
from transformers import PreTrainedTokenizerBase, PretrainedConfig
Fix typing in `Model.generate_token` (#733) ## What does this PR do? This PR fixes a minor type annotation issue in the signature of `Model.generate_token`. All existing overrides of `Model.generate_token` return `Tuple[List[Generation], Optional[B]]`: https://github.com/huggingface/text-generation-inference/blob/3ef5ffbc6400370ff2e1546550a6bad3ac61b079/server/text_generation_server/models/causal_lm.py#L535-L537 https://github.com/huggingface/text-generation-inference/blob/3ef5ffbc6400370ff2e1546550a6bad3ac61b079/server/text_generation_server/models/flash_causal_lm.py#L802-L804 https://github.com/huggingface/text-generation-inference/blob/3ef5ffbc6400370ff2e1546550a6bad3ac61b079/server/text_generation_server/models/seq2seq_lm.py#L589-L591 I suspect that back in 017a2a8c when `GeneratedText` and `Generation` were separated, the function signature was not updated. ## Before submitting - [ ] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). - [x] Did you read the [contributor guideline](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md#start-contributing-pull-requests), Pull Request section? - [ ] Was this discussed/approved via a Github issue or the [forum](https://discuss.huggingface.co/)? Please add a link to it if that's the case. - [ ] Did you make sure to update the documentation with your changes? Here are the [documentation guidelines](https://github.com/huggingface/transformers/tree/main/docs), and [here are tips on formatting docstrings](https://github.com/huggingface/transformers/tree/main/docs#writing-source-documentation). - [ ] Did you write any new necessary tests? CC @OlivierDehaene
2023-07-31 06:35:14 -06:00
from text_generation_server.models.types import Batch, Generation
2023-12-11 04:46:30 -07:00
from text_generation_server.utils.speculate import get_speculate
from text_generation_server.pb.generate_pb2 import InfoResponse
B = TypeVar("B", bound=Batch)
class Model(ABC):
def __init__(
self,
model: torch.nn.Module,
tokenizer: PreTrainedTokenizerBase,
requires_padding: bool,
dtype: torch.dtype,
device: torch.device,
rank: int = 0,
world_size: int = 1,
2023-09-28 01:55:47 -06:00
sliding_window: Optional[int] = None,
2023-12-11 04:46:30 -07:00
speculate: Optional[int] = None,
):
self.model = model.eval()
self.tokenizer = tokenizer
self.all_special_ids = set(tokenizer.all_special_ids)
self.requires_padding = requires_padding
self.dtype = dtype
self.device = device
self.rank = rank
self.world_size = world_size
self.sliding_window = sliding_window if sliding_window != -1 else None
2023-12-11 04:46:30 -07:00
if speculate is None:
speculate = get_speculate()
self.speculate = speculate
self.has_position_ids = (
inspect.signature(model.forward).parameters.get("position_ids", None)
is not None
)
Lifting check_unitialized. (#325) # What does this PR do? Lifting check_unitialized. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflects the extent of your awesome contribution. Then, please replace this with a description of the change and which issue is fixed (if applicable). Please also include relevant motivation and context. List any dependencies (if any) that are required for this change. Once you're done, someone will review your PR shortly (see the section "Who can review?" below to tag some potential reviewers). They may suggest changes to make the code even better. If no one reviewed your PR after a week has passed, don't hesitate to post a new comment @-mentioning the same persons---sometimes notifications get lost. --> <!-- Remove if not applicable --> Fixes # (issue) ## Before submitting - [ ] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). - [ ] Did you read the [contributor guideline](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md#start-contributing-pull-requests), Pull Request section? - [ ] Was this discussed/approved via a Github issue or the [forum](https://discuss.huggingface.co/)? Please add a link to it if that's the case. - [ ] Did you make sure to update the documentation with your changes? Here are the [documentation guidelines](https://github.com/huggingface/transformers/tree/main/docs), and [here are tips on formatting docstrings](https://github.com/huggingface/transformers/tree/main/docs#writing-source-documentation). - [ ] Did you write any new necessary tests? ## Who can review? Anyone in the community is free to review the PR once the tests have passed. Feel free to tag members/contributors who may be interested in your PR. <!-- Your PR will be replied to more quickly if you can figure out the right person to tag with @ @OlivierDehaene OR @Narsil -->
2023-05-15 03:32:25 -06:00
self.check_initialized()
@property
def info(self) -> InfoResponse:
2023-09-28 01:55:47 -06:00
if self.requires_padding and self.sliding_window is not None:
raise NotImplementedError("sliding_window is not implemented with padding")
return InfoResponse(
requires_padding=self.requires_padding,
dtype=str(self.dtype),
device_type=self.device.type,
2023-09-28 01:55:47 -06:00
window_size=self.sliding_window,
2023-12-11 06:49:52 -07:00
speculate=self.speculate,
)
@property
@abstractmethod
def batch_type(self) -> Type[B]:
raise NotImplementedError
@abstractmethod
def generate_token(
self, batch: B
) -> Tuple[List[Generation], Optional[B], Tuple[int, int]]:
raise NotImplementedError
def warmup(self, batch: B) -> Optional[int]:
self.generate_token(batch)
return None
def decode_token(
self,
all_input_ids: List[int],
prefix_offset: int = 0,
read_offset: int = 0,
Remove the stripping of the prefix space (and any other mangling that tokenizers might do). (#1065) Superseed #1024 # What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflects the extent of your awesome contribution. Then, please replace this with a description of the change and which issue is fixed (if applicable). Please also include relevant motivation and context. List any dependencies (if any) that are required for this change. Once you're done, someone will review your PR shortly (see the section "Who can review?" below to tag some potential reviewers). They may suggest changes to make the code even better. If no one reviewed your PR after a week has passed, don't hesitate to post a new comment @-mentioning the same persons---sometimes notifications get lost. --> <!-- Remove if not applicable --> Fixes # (issue) ## Before submitting - [ ] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). - [ ] Did you read the [contributor guideline](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md#start-contributing-pull-requests), Pull Request section? - [ ] Was this discussed/approved via a Github issue or the [forum](https://discuss.huggingface.co/)? Please add a link to it if that's the case. - [ ] Did you make sure to update the documentation with your changes? Here are the [documentation guidelines](https://github.com/huggingface/transformers/tree/main/docs), and [here are tips on formatting docstrings](https://github.com/huggingface/transformers/tree/main/docs#writing-source-documentation). - [ ] Did you write any new necessary tests? ## Who can review? Anyone in the community is free to review the PR once the tests have passed. Feel free to tag members/contributors who may be interested in your PR. <!-- Your PR will be replied to more quickly if you can figure out the right person to tag with @ @OlivierDehaene OR @Narsil --> --------- Co-authored-by: bangoz <ch_xie@pku.edu.cn>
2023-09-27 04:13:45 -06:00
skip_special_tokens: bool = False,
) -> Tuple[str, int, int]:
"""Hack to hopefully support generate_stream for the maximum number of tokenizers"""
# The prefix text is necessary only to defeat cleanup algorithms in the decode
# which decide to add a space or not depending on the surrounding ids.
prefix_text = self.tokenizer.decode(
2023-09-27 04:22:09 -06:00
all_input_ids[prefix_offset:read_offset],
skip_special_tokens=skip_special_tokens,
)
new_text = self.tokenizer.decode(
Remove the stripping of the prefix space (and any other mangling that tokenizers might do). (#1065) Superseed #1024 # What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflects the extent of your awesome contribution. Then, please replace this with a description of the change and which issue is fixed (if applicable). Please also include relevant motivation and context. List any dependencies (if any) that are required for this change. Once you're done, someone will review your PR shortly (see the section "Who can review?" below to tag some potential reviewers). They may suggest changes to make the code even better. If no one reviewed your PR after a week has passed, don't hesitate to post a new comment @-mentioning the same persons---sometimes notifications get lost. --> <!-- Remove if not applicable --> Fixes # (issue) ## Before submitting - [ ] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). - [ ] Did you read the [contributor guideline](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md#start-contributing-pull-requests), Pull Request section? - [ ] Was this discussed/approved via a Github issue or the [forum](https://discuss.huggingface.co/)? Please add a link to it if that's the case. - [ ] Did you make sure to update the documentation with your changes? Here are the [documentation guidelines](https://github.com/huggingface/transformers/tree/main/docs), and [here are tips on formatting docstrings](https://github.com/huggingface/transformers/tree/main/docs#writing-source-documentation). - [ ] Did you write any new necessary tests? ## Who can review? Anyone in the community is free to review the PR once the tests have passed. Feel free to tag members/contributors who may be interested in your PR. <!-- Your PR will be replied to more quickly if you can figure out the right person to tag with @ @OlivierDehaene OR @Narsil --> --------- Co-authored-by: bangoz <ch_xie@pku.edu.cn>
2023-09-27 04:13:45 -06:00
all_input_ids[prefix_offset:], skip_special_tokens=skip_special_tokens
)
if len(new_text) > len(prefix_text) and not new_text.endswith("<EFBFBD>"):
# utf-8 char at the end means it's a potential unfinished byte sequence
# from byte fallback tokenization.
# If it's in the middle, it's probably a real invalid id generated
# by the model
new_text = new_text[len(prefix_text) :]
return new_text, read_offset, len(all_input_ids)
else:
return "", prefix_offset, read_offset
Lifting check_unitialized. (#325) # What does this PR do? Lifting check_unitialized. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflects the extent of your awesome contribution. Then, please replace this with a description of the change and which issue is fixed (if applicable). Please also include relevant motivation and context. List any dependencies (if any) that are required for this change. Once you're done, someone will review your PR shortly (see the section "Who can review?" below to tag some potential reviewers). They may suggest changes to make the code even better. If no one reviewed your PR after a week has passed, don't hesitate to post a new comment @-mentioning the same persons---sometimes notifications get lost. --> <!-- Remove if not applicable --> Fixes # (issue) ## Before submitting - [ ] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). - [ ] Did you read the [contributor guideline](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md#start-contributing-pull-requests), Pull Request section? - [ ] Was this discussed/approved via a Github issue or the [forum](https://discuss.huggingface.co/)? Please add a link to it if that's the case. - [ ] Did you make sure to update the documentation with your changes? Here are the [documentation guidelines](https://github.com/huggingface/transformers/tree/main/docs), and [here are tips on formatting docstrings](https://github.com/huggingface/transformers/tree/main/docs#writing-source-documentation). - [ ] Did you write any new necessary tests? ## Who can review? Anyone in the community is free to review the PR once the tests have passed. Feel free to tag members/contributors who may be interested in your PR. <!-- Your PR will be replied to more quickly if you can figure out the right person to tag with @ @OlivierDehaene OR @Narsil -->
2023-05-15 03:32:25 -06:00
def check_initialized(self):
uninitialized_parameters = []
for n, p in self.model.named_parameters():
if p.data.device == torch.device("meta"):
uninitialized_parameters.append(n)
if uninitialized_parameters:
raise RuntimeError(
f"found uninitialized parameters in model {self.__class__.__name__}: {uninitialized_parameters}"
)