ef6abf3dc7
Adds ALL files needed for Dockerfile build:
- qwen3_6_scripts/ (baseline patches + our optimizations)
- vllm/ (full vllm package)
- paged_attention_v2_pytorch.py (V2 with single-bmm optimization)
- Dockerfile + computility-run.yaml
Our optimizations vs baseline:
1. paged_attn.py: pre-gathered context KV (eliminates 194 gather calls),
Triton try/fallback, V2 heuristic, threshold 32K→64K
2. paged_attention_v2_pytorch.py: fills NotImplementedError,
single-bmm Phase 1 (195 launches → 3)
3. patch_enable_triton.py: HAS_TRITON=True with safety fallback
4. patch_triton_tuning.py: BLOCK=64, NUM_WARPS=4 for BI-V100
5. computility-run.yaml: gpu-memory-utilization 0.9→0.95,
max-num-batched-tokens 8192→16384
This repo can now be submitted to dev.modelhub.org.cn as-is.
33 lines
1.2 KiB
Python
33 lines
1.2 KiB
Python
from typing import Any, Dict, Optional, Union
|
|
|
|
import torch
|
|
import torch.distributed
|
|
|
|
from .parallel_state import get_tp_group
|
|
|
|
|
|
def tensor_model_parallel_all_reduce(input_: torch.Tensor) -> torch.Tensor:
|
|
"""All-reduce the input tensor across model parallel group."""
|
|
return get_tp_group().all_reduce(input_)
|
|
|
|
|
|
def tensor_model_parallel_all_gather(input_: torch.Tensor,
|
|
dim: int = -1) -> torch.Tensor:
|
|
"""All-gather the input tensor across model parallel group."""
|
|
return get_tp_group().all_gather(input_, dim)
|
|
|
|
|
|
def tensor_model_parallel_gather(input_: torch.Tensor,
|
|
dst: int = 0,
|
|
dim: int = -1) -> Optional[torch.Tensor]:
|
|
"""Gather the input tensor across model parallel group."""
|
|
return get_tp_group().gather(input_, dst, dim)
|
|
|
|
|
|
def broadcast_tensor_dict(tensor_dict: Optional[Dict[Any, Union[torch.Tensor,
|
|
Any]]] = None,
|
|
src: int = 0):
|
|
if not torch.distributed.is_initialized():
|
|
return tensor_dict
|
|
return get_tp_group().broadcast_tensor_dict(tensor_dict, src)
|