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.
25 lines
675 B
Python
25 lines
675 B
Python
from .base import (BatchedTensorInputs, MultiModalDataBuiltins,
|
|
MultiModalDataDict, MultiModalInputs, MultiModalPlugin,
|
|
NestedTensors)
|
|
from .registry import MultiModalRegistry
|
|
|
|
MULTIMODAL_REGISTRY = MultiModalRegistry()
|
|
"""
|
|
The global :class:`~MultiModalRegistry` is used by model runners to
|
|
dispatch data processing according to its modality and the target model.
|
|
|
|
See also:
|
|
:ref:`input_processing_pipeline`
|
|
"""
|
|
|
|
__all__ = [
|
|
"BatchedTensorInputs",
|
|
"MultiModalDataBuiltins",
|
|
"MultiModalDataDict",
|
|
"MultiModalInputs",
|
|
"MultiModalPlugin",
|
|
"NestedTensors",
|
|
"MULTIMODAL_REGISTRY",
|
|
"MultiModalRegistry",
|
|
]
|