mirror of
https://github.com/huggingface/text-generation-inference.git
synced 2025-04-21 23:12:07 +00:00
a072660bf5
# What does this PR do? A few tokenizer_config in huggingface use LlamaTokenizer, so I think I would have selected `LlamaTokenizer` before. For a few cases where you're using a llama structure but not a llama tokenizer, why not make it to call the AutoTokenizer in exception handling. In the case of `decapoda-research/llama-7b-hf`, LLamaTokenizer is still being used in config.json, so it should be called through` LlamaTokenizer`. Also, if an exception is thrown by LlamaTokenizer, it will cause `LlamaTokenzierFast` to be called from AutoTokenizer. Fixes # 560 ## 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. - [x] 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. @Narsil
82 lines
2.6 KiB
Python
82 lines
2.6 KiB
Python
import torch
|
|
import torch.distributed
|
|
|
|
from opentelemetry import trace
|
|
from transformers import AutoConfig, AutoTokenizer
|
|
from transformers.models.llama import LlamaTokenizer
|
|
from typing import Optional
|
|
|
|
from text_generation_server.models import FlashCausalLM
|
|
from text_generation_server.models.custom_modeling.flash_llama_modeling import (
|
|
FlashLlamaForCausalLM,
|
|
LlamaConfig,
|
|
)
|
|
from text_generation_server.utils import (
|
|
initialize_torch_distributed,
|
|
weight_files,
|
|
Weights,
|
|
)
|
|
|
|
tracer = trace.get_tracer(__name__)
|
|
|
|
|
|
class FlashLlama(FlashCausalLM):
|
|
def __init__(
|
|
self,
|
|
model_id: str,
|
|
revision: Optional[str] = None,
|
|
quantize: Optional[str] = None,
|
|
dtype: Optional[torch.dtype] = None,
|
|
trust_remote_code: bool = False,
|
|
):
|
|
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
|
|
else:
|
|
raise NotImplementedError("FlashLlama is only available on GPU")
|
|
|
|
try:
|
|
tokenizer = LlamaTokenizer.from_pretrained(
|
|
model_id,
|
|
revision=revision,
|
|
padding_side="left",
|
|
truncation_side="left",
|
|
trust_remote_code=trust_remote_code,
|
|
)
|
|
except Exception:
|
|
tokenizer = AutoTokenizer.from_pretrained(
|
|
model_id,
|
|
revision=revision,
|
|
padding_side="left",
|
|
truncation_side="left",
|
|
trust_remote_code=trust_remote_code,
|
|
)
|
|
|
|
config = LlamaConfig.from_pretrained(
|
|
model_id, revision=revision, trust_remote_code=trust_remote_code
|
|
)
|
|
config.quantize = quantize
|
|
|
|
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 == "gptq":
|
|
weights._set_gptq_params(model_id)
|
|
|
|
model = FlashLlamaForCausalLM(config, weights)
|
|
|
|
torch.distributed.barrier(group=self.process_group)
|
|
super(FlashLlama, self).__init__(
|
|
model=model,
|
|
tokenizer=tokenizer,
|
|
num_layers=len(model.model.layers),
|
|
num_kv_heads=model.model.num_key_value_heads,
|
|
head_size=model.model.head_size,
|
|
dtype=dtype,
|
|
device=device,
|
|
rank=rank,
|
|
world_size=world_size,
|
|
)
|