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

122 lines
3.9 KiB
Python
Raw Normal View History

2023-09-28 01:55:47 -06:00
import torch
import torch.distributed
from opentelemetry import trace
from transformers import AutoTokenizer, AutoConfig
from typing import Optional, Tuple
2023-09-28 01:55:47 -06:00
from text_generation_server.models import FlashCausalLM
from text_generation_server.models.flash_causal_lm import set_sliding_window
2023-09-28 01:55:47 -06:00
from text_generation_server.models.custom_modeling.flash_mistral_modeling import (
FlashMistralForCausalLM,
MistralConfig,
)
from text_generation_server.utils import (
initialize_torch_distributed,
weight_files,
Weights,
)
Refactor layers. (#1866) # 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 -->
2024-05-13 04:44:30 -06:00
from text_generation_server.utils.import_utils import SYSTEM
2023-09-28 01:55:47 -06:00
tracer = trace.get_tracer(__name__)
2023-09-28 01:55:47 -06:00
2023-12-11 06:43:40 -07:00
class BaseFlashMistral(FlashCausalLM):
2023-09-28 01:55:47 -06:00
def __init__(
2023-12-11 06:49:52 -07:00
self,
model_cls,
model_id: str,
Adding Llava-Next (Llava 1.6) with full support. (#1709) # What does this PR do? - Changed all models to extract `embed_tokens` in order to enable llava to separately call the embeddings and the core model layers. - Added VlmCausalLM to inherit from FlashMistral in order to be maximally supported. The only added logics sits on top and parses images into pixel values, preallocates input_ids space for the image embeddings, and passes them for the model. - Added Clip for the vision tower. - Didn't add flash for the vision tower since there's no padding anyway. - Added heuristic (potentially incomplete) to calculate number of features *before* calculating the clip patches (allows for easier logic reuse of the LLM under the hood). Still needs to be done: - [x] Implement the image parsing in the controller side, to avoid downloading n times per TP shard and also refusing requests too large early and avoid issues where the truncation actually truncates the image. - [ ] Make sure it works with quantization properly. - [x] Make sure it works with TP>1 <!-- 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 -->
2024-04-09 13:32:00 -06:00
config_cls=AutoConfig,
2023-12-11 06:49:52 -07:00
revision: Optional[str] = None,
quantize: Optional[str] = None,
MLPSpeculator. (#1865) # 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: Joshua Rosenkranz <joshua.rosenkranz@gmail.com>
2024-05-14 04:33:18 -06:00
speculator: Optional[str] = None,
2023-12-11 06:49:52 -07:00
dtype: Optional[torch.dtype] = None,
trust_remote_code: bool = False,
Adding Llava-Next (Llava 1.6) with full support. (#1709) # What does this PR do? - Changed all models to extract `embed_tokens` in order to enable llava to separately call the embeddings and the core model layers. - Added VlmCausalLM to inherit from FlashMistral in order to be maximally supported. The only added logics sits on top and parses images into pixel values, preallocates input_ids space for the image embeddings, and passes them for the model. - Added Clip for the vision tower. - Didn't add flash for the vision tower since there's no padding anyway. - Added heuristic (potentially incomplete) to calculate number of features *before* calculating the clip patches (allows for easier logic reuse of the LLM under the hood). Still needs to be done: - [x] Implement the image parsing in the controller side, to avoid downloading n times per TP shard and also refusing requests too large early and avoid issues where the truncation actually truncates the image. - [ ] Make sure it works with quantization properly. - [x] Make sure it works with TP>1 <!-- 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 -->
2024-04-09 13:32:00 -06:00
tokenizer_class=AutoTokenizer,
2023-09-28 01:55:47 -06:00
):
self.process_group, rank, world_size = initialize_torch_distributed()
if torch.cuda.is_available():
device = torch.device(f"cuda:{rank}")
dtype = torch.float16 if dtype is None else dtype
Refactor layers. (#1866) # 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 -->
2024-05-13 04:44:30 -06:00
elif SYSTEM == "xpu":
add intel xpu support for TGI (#1475) # 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 --> --------- Signed-off-by: Wang, Yi A <yi.a.wang@intel.com> Co-authored-by: Morgan Funtowicz <funtowiczmo@gmail.com> Co-authored-by: Nicolas Patry <patry.nicolas@protonmail.com>
2024-04-26 07:48:58 -06:00
device = torch.device(f"xpu:{rank}")
dtype = torch.float16 if dtype is None else dtype
2023-09-28 01:55:47 -06:00
else:
raise NotImplementedError("FlashMistral is only available on GPU")
2023-09-28 01:55:47 -06:00
Adding Llava-Next (Llava 1.6) with full support. (#1709) # What does this PR do? - Changed all models to extract `embed_tokens` in order to enable llava to separately call the embeddings and the core model layers. - Added VlmCausalLM to inherit from FlashMistral in order to be maximally supported. The only added logics sits on top and parses images into pixel values, preallocates input_ids space for the image embeddings, and passes them for the model. - Added Clip for the vision tower. - Didn't add flash for the vision tower since there's no padding anyway. - Added heuristic (potentially incomplete) to calculate number of features *before* calculating the clip patches (allows for easier logic reuse of the LLM under the hood). Still needs to be done: - [x] Implement the image parsing in the controller side, to avoid downloading n times per TP shard and also refusing requests too large early and avoid issues where the truncation actually truncates the image. - [ ] Make sure it works with quantization properly. - [x] Make sure it works with TP>1 <!-- 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 -->
2024-04-09 13:32:00 -06:00
tokenizer = tokenizer_class.from_pretrained(
model_id,
revision=revision,
padding_side="left",
truncation_side="left",
trust_remote_code=trust_remote_code,
)
2023-09-28 01:55:47 -06:00
2023-12-11 06:43:40 -07:00
config = config_cls.from_pretrained(
2023-09-28 01:55:47 -06:00
model_id, revision=revision, trust_remote_code=trust_remote_code
)
config.quantize = quantize
MLPSpeculator. (#1865) # 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: Joshua Rosenkranz <joshua.rosenkranz@gmail.com>
2024-05-14 04:33:18 -06:00
config.speculator = speculator
2023-09-28 01:55:47 -06:00
# Set context windows
Adding Llava-Next (Llava 1.6) with full support. (#1709) # What does this PR do? - Changed all models to extract `embed_tokens` in order to enable llava to separately call the embeddings and the core model layers. - Added VlmCausalLM to inherit from FlashMistral in order to be maximally supported. The only added logics sits on top and parses images into pixel values, preallocates input_ids space for the image embeddings, and passes them for the model. - Added Clip for the vision tower. - Didn't add flash for the vision tower since there's no padding anyway. - Added heuristic (potentially incomplete) to calculate number of features *before* calculating the clip patches (allows for easier logic reuse of the LLM under the hood). Still needs to be done: - [x] Implement the image parsing in the controller side, to avoid downloading n times per TP shard and also refusing requests too large early and avoid issues where the truncation actually truncates the image. - [ ] Make sure it works with quantization properly. - [x] Make sure it works with TP>1 <!-- 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 -->
2024-04-09 13:32:00 -06:00
if getattr(config, "sliding_window", None) is not None:
set_sliding_window(config.sliding_window)
Adding Llava-Next (Llava 1.6) with full support. (#1709) # What does this PR do? - Changed all models to extract `embed_tokens` in order to enable llava to separately call the embeddings and the core model layers. - Added VlmCausalLM to inherit from FlashMistral in order to be maximally supported. The only added logics sits on top and parses images into pixel values, preallocates input_ids space for the image embeddings, and passes them for the model. - Added Clip for the vision tower. - Didn't add flash for the vision tower since there's no padding anyway. - Added heuristic (potentially incomplete) to calculate number of features *before* calculating the clip patches (allows for easier logic reuse of the LLM under the hood). Still needs to be done: - [x] Implement the image parsing in the controller side, to avoid downloading n times per TP shard and also refusing requests too large early and avoid issues where the truncation actually truncates the image. - [ ] Make sure it works with quantization properly. - [x] Make sure it works with TP>1 <!-- 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 -->
2024-04-09 13:32:00 -06:00
else:
config.sliding_window = None
2023-09-28 01:55:47 -06:00
torch.distributed.barrier(group=self.process_group)
filenames = weight_files(model_id, revision=revision, extension=".safetensors")
weights = Weights(filenames, device, dtype, process_group=self.process_group)
if config.quantize in ["gptq", "awq", "marlin"]:
2023-12-14 03:02:16 -07:00
weights._set_gptq_params(model_id, revision)
2023-09-28 01:55:47 -06:00
Adding Llava-Next (Llava 1.6) with full support. (#1709) # What does this PR do? - Changed all models to extract `embed_tokens` in order to enable llava to separately call the embeddings and the core model layers. - Added VlmCausalLM to inherit from FlashMistral in order to be maximally supported. The only added logics sits on top and parses images into pixel values, preallocates input_ids space for the image embeddings, and passes them for the model. - Added Clip for the vision tower. - Didn't add flash for the vision tower since there's no padding anyway. - Added heuristic (potentially incomplete) to calculate number of features *before* calculating the clip patches (allows for easier logic reuse of the LLM under the hood). Still needs to be done: - [x] Implement the image parsing in the controller side, to avoid downloading n times per TP shard and also refusing requests too large early and avoid issues where the truncation actually truncates the image. - [ ] Make sure it works with quantization properly. - [x] Make sure it works with TP>1 <!-- 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 -->
2024-04-09 13:32:00 -06:00
prefix = ""
model = model_cls(prefix, config, weights)
2023-09-28 01:55:47 -06:00
self.cuda_graphs = {}
2023-09-28 01:55:47 -06:00
torch.distributed.barrier(group=self.process_group)
Adding Llava-Next (Llava 1.6) with full support. (#1709) # What does this PR do? - Changed all models to extract `embed_tokens` in order to enable llava to separately call the embeddings and the core model layers. - Added VlmCausalLM to inherit from FlashMistral in order to be maximally supported. The only added logics sits on top and parses images into pixel values, preallocates input_ids space for the image embeddings, and passes them for the model. - Added Clip for the vision tower. - Didn't add flash for the vision tower since there's no padding anyway. - Added heuristic (potentially incomplete) to calculate number of features *before* calculating the clip patches (allows for easier logic reuse of the LLM under the hood). Still needs to be done: - [x] Implement the image parsing in the controller side, to avoid downloading n times per TP shard and also refusing requests too large early and avoid issues where the truncation actually truncates the image. - [ ] Make sure it works with quantization properly. - [x] Make sure it works with TP>1 <!-- 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 -->
2024-04-09 13:32:00 -06:00
num_layers, num_kv_heads, head_size = self.get_layer_config(model)
super().__init__(
2023-09-28 01:55:47 -06:00
model=model,
tokenizer=tokenizer,
Adding Llava-Next (Llava 1.6) with full support. (#1709) # What does this PR do? - Changed all models to extract `embed_tokens` in order to enable llava to separately call the embeddings and the core model layers. - Added VlmCausalLM to inherit from FlashMistral in order to be maximally supported. The only added logics sits on top and parses images into pixel values, preallocates input_ids space for the image embeddings, and passes them for the model. - Added Clip for the vision tower. - Didn't add flash for the vision tower since there's no padding anyway. - Added heuristic (potentially incomplete) to calculate number of features *before* calculating the clip patches (allows for easier logic reuse of the LLM under the hood). Still needs to be done: - [x] Implement the image parsing in the controller side, to avoid downloading n times per TP shard and also refusing requests too large early and avoid issues where the truncation actually truncates the image. - [ ] Make sure it works with quantization properly. - [x] Make sure it works with TP>1 <!-- 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 -->
2024-04-09 13:32:00 -06:00
num_layers=num_layers,
num_kv_heads=num_kv_heads,
head_size=head_size,
2023-09-28 01:55:47 -06:00
dtype=dtype,
device=device,
rank=rank,
world_size=world_size,
sliding_window=config.sliding_window,
)
Adding Llava-Next (Llava 1.6) with full support. (#1709) # What does this PR do? - Changed all models to extract `embed_tokens` in order to enable llava to separately call the embeddings and the core model layers. - Added VlmCausalLM to inherit from FlashMistral in order to be maximally supported. The only added logics sits on top and parses images into pixel values, preallocates input_ids space for the image embeddings, and passes them for the model. - Added Clip for the vision tower. - Didn't add flash for the vision tower since there's no padding anyway. - Added heuristic (potentially incomplete) to calculate number of features *before* calculating the clip patches (allows for easier logic reuse of the LLM under the hood). Still needs to be done: - [x] Implement the image parsing in the controller side, to avoid downloading n times per TP shard and also refusing requests too large early and avoid issues where the truncation actually truncates the image. - [ ] Make sure it works with quantization properly. - [x] Make sure it works with TP>1 <!-- 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 -->
2024-04-09 13:32:00 -06:00
def get_layer_config(self, model) -> Tuple[int, int, int]:
return (
len(model.model.layers),
model.model.num_key_value_heads,
model.model.head_size,
)
2023-12-11 06:43:40 -07:00
class FlashMistral(BaseFlashMistral):
def __init__(
2023-12-11 06:49:52 -07:00
self,
model_id: str,
revision: Optional[str] = None,
quantize: Optional[str] = None,
MLPSpeculator. (#1865) # 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: Joshua Rosenkranz <joshua.rosenkranz@gmail.com>
2024-05-14 04:33:18 -06:00
speculator: Optional[str] = None,
2023-12-11 06:49:52 -07:00
dtype: Optional[torch.dtype] = None,
trust_remote_code: bool = False,
2023-12-11 06:43:40 -07:00
):
super(FlashMistral, self).__init__(
config_cls=MistralConfig,
model_cls=FlashMistralForCausalLM,
model_id=model_id,
revision=revision,
quantize=quantize,
MLPSpeculator. (#1865) # 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: Joshua Rosenkranz <joshua.rosenkranz@gmail.com>
2024-05-14 04:33:18 -06:00
speculator=speculator,
2023-12-11 06:43:40 -07:00
dtype=dtype,
2023-12-11 06:49:52 -07:00
trust_remote_code=trust_remote_code,
2023-12-11 06:43:40 -07:00
)