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.
24 lines
549 B
Python
24 lines
549 B
Python
from contextlib import contextmanager
|
|
from typing import Any
|
|
|
|
_compile_context: Any = None
|
|
|
|
|
|
def get_compile_context() -> Any:
|
|
"""Get the current compile context."""
|
|
return _compile_context
|
|
|
|
|
|
@contextmanager
|
|
def set_compile_context(context: Any):
|
|
"""A context manager that stores the current compile context,
|
|
usually it is a list of sizes to specialize.
|
|
"""
|
|
global _compile_context
|
|
prev_context = _compile_context
|
|
_compile_context = context
|
|
try:
|
|
yield
|
|
finally:
|
|
_compile_context = prev_context
|