mirror of
https://github.com/huggingface/text-generation-inference.git
synced 2025-04-22 15:32:08 +00:00
b70ae0969f
* Prefix caching WIP * Fixing prefix attention. * Fixing flashinfer import. * Fixing black. * Fixing medusa (still wrong outputs, but functional). * Just medusa values now. * Fixing medusa without prefix caching. * Fixing prefix caching. * Medusa requires reshaping. * Removing the logs. * Remove router.nix * Fixup: - Remove logs - Disable VLMs (they do not work) - Disable prefix caching when user wants prefill logprobs. * Update flake.lock --------- Co-authored-by: Daniël de Kok <me@danieldk.eu>
30 lines
796 B
Python
30 lines
796 B
Python
from text_generation_server.utils.import_utils import SYSTEM
|
|
import os
|
|
|
|
from .common import Seqlen
|
|
|
|
if os.getenv("USE_FLASH_ATTENTION", "").lower() == "false":
|
|
raise ImportError("`USE_FLASH_ATTENTION` is false.")
|
|
if SYSTEM == "cuda":
|
|
from .cuda import (
|
|
attention,
|
|
paged_attention,
|
|
reshape_and_cache,
|
|
SUPPORTS_WINDOWING,
|
|
)
|
|
elif SYSTEM == "rocm":
|
|
from .rocm import attention, paged_attention, reshape_and_cache, SUPPORTS_WINDOWING
|
|
elif SYSTEM == "ipex":
|
|
from .ipex import attention, paged_attention, reshape_and_cache, SUPPORTS_WINDOWING
|
|
else:
|
|
raise ImportError(f"System {SYSTEM} doesn't support flash/paged attention")
|
|
|
|
|
|
__all__ = [
|
|
"attention",
|
|
"paged_attention",
|
|
"reshape_and_cache",
|
|
"SUPPORTS_WINDOWING",
|
|
"Seqlen",
|
|
]
|