mirror of
https://github.com/kvcache-ai/ktransformers.git
synced 2026-04-28 11:49:51 +00:00
57d14d22bc
* refactor: move legacy code to archive/ directory - Moved ktransformers, csrc, third_party, merge_tensors to archive/ - Moved build scripts and configurations to archive/ - Kept kt-kernel, KT-SFT, doc, and README files in root - Preserved complete git history for all moved files * refactor: restructure repository to focus on kt-kernel and KT-SFT modules * fix README * fix README * fix README * fix README * docs: add performance benchmarks to kt-kernel section Add comprehensive performance data for kt-kernel to match KT-SFT's presentation: - AMX kernel optimization: 21.3 TFLOPS (3.9× faster than PyTorch) - Prefill phase: up to 20× speedup vs baseline - Decode phase: up to 4× speedup - NUMA optimization: up to 63% throughput improvement - Multi-GPU (8×L20): 227.85 tokens/s total throughput with DeepSeek-R1 FP8 Source: https://lmsys.org/blog/2025-10-22-KTransformers/ This provides users with concrete performance metrics for both core modules, making it easier to understand the capabilities of each component. * refactor: improve kt-kernel performance data with specific hardware and models Replace generic performance descriptions with concrete benchmarks: - Specify exact hardware: 8×L20 GPU + Xeon Gold 6454S, Single/Dual-socket Xeon + AMX - Include specific models: DeepSeek-R1-0528 (FP8), DeepSeek-V3 (671B) - Show detailed metrics: total throughput, output throughput, concurrency details - Match KT-SFT presentation style for consistency This provides users with actionable performance data they can use to evaluate hardware requirements and expected performance for their use cases. * fix README * docs: clean up performance table and improve formatting * add pic for README * refactor: simplify .gitmodules and backup legacy submodules - Remove 7 legacy submodules from root .gitmodules (archive/third_party/*) - Keep only 2 active submodules for kt-kernel (llama.cpp, pybind11) - Backup complete .gitmodules to archive/.gitmodules - Add documentation in archive/README.md for researchers who need legacy submodules This reduces initial clone size by ~500MB and avoids downloading unused dependencies. * refactor: move doc/ back to root directory Keep documentation in root for easier access and maintenance. * refactor: consolidate all images to doc/assets/ - Move kt-kernel/assets/heterogeneous_computing.png to doc/assets/ - Remove KT-SFT/assets/ (images already in doc/assets/) - Update KT-SFT/README.md image references to ../doc/assets/ - Eliminates ~7.9MB image duplication - Centralizes all documentation assets in one location * fix pic path for README
65 lines
No EOL
3.2 KiB
Python
65 lines
No EOL
3.2 KiB
Python
from asyncio import Lock
|
|
from typing import Dict, Optional
|
|
|
|
from ktransformers.server.backend.base import ThreadContext, BackendInterfaceBase
|
|
from ktransformers.server.schemas.assistants.runs import RunObject
|
|
from ktransformers.server.schemas.base import ObjectID
|
|
from ktransformers.server.config.log import logger
|
|
from ktransformers.server.backend.interfaces.transformers import TransformersThreadContext
|
|
from ktransformers.server.backend.interfaces.ktransformers import KTransformersThreadContext
|
|
from ktransformers.server.backend.interfaces.exllamav2 import ExllamaThreadContext
|
|
|
|
|
|
from ktransformers.server.backend.interfaces.exllamav2 import ExllamaInterface
|
|
from ktransformers.server.backend.interfaces.transformers import TransformersInterface
|
|
from ktransformers.server.backend.interfaces.ktransformers import KTransformersInterface
|
|
|
|
class ThreadContextManager:
|
|
lock: Lock
|
|
threads_context: Dict[ObjectID, ThreadContext]
|
|
interface: BackendInterfaceBase
|
|
|
|
def __init__(self,interface) -> None:
|
|
logger.debug(f"Creating Context Manager")
|
|
self.lock = Lock()
|
|
self.threads_context = {}
|
|
self.interface = interface
|
|
pass
|
|
|
|
async def get_context_by_run_object(self, run: RunObject) -> ThreadContext:
|
|
async with self.lock:
|
|
logger.debug(f"keys {self.threads_context.keys()}")
|
|
if run.thread_id not in self.threads_context:
|
|
logger.debug(f"new inference context {run.thread_id}")
|
|
if isinstance(self.interface, ExllamaInterface):
|
|
new_context = ExllamaThreadContext(run, self.interface)
|
|
elif isinstance(self.interface, KTransformersInterface):
|
|
new_context = KTransformersThreadContext(run, self.interface)
|
|
elif isinstance(self.interface, TransformersInterface):
|
|
new_context = TransformersThreadContext(run, self.interface)
|
|
else:
|
|
from ktransformers.server.backend.interfaces.balance_serve import BalanceServeThreadContext
|
|
from ktransformers.server.backend.interfaces.balance_serve import BalanceServeInterface
|
|
if isinstance(self.interface, BalanceServeInterface):
|
|
new_context = BalanceServeThreadContext(run, self.interface)
|
|
else:
|
|
raise NotImplementedError
|
|
# elif isinstance(self.interface, BalanceServeInterface):
|
|
# new_context = BalanceServeThreadContext(run, self.interface)
|
|
# else:
|
|
# raise NotImplementedError
|
|
self.threads_context[run.thread_id] = new_context
|
|
# self.threads_context[run.thread_id] = ExllamaInferenceContext(run)
|
|
re = self.threads_context[run.thread_id]
|
|
re.update_by_run(run)
|
|
return re
|
|
|
|
async def get_context_by_thread_id(self, thread_id: ObjectID) -> Optional[ThreadContext]:
|
|
async with self.lock:
|
|
if thread_id in self.threads_context:
|
|
logger.debug(f'found context for thread {thread_id}')
|
|
return self.threads_context[thread_id]
|
|
else:
|
|
logger.debug(f'no context for thread {thread_id}')
|
|
return None
|
|
|