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.
29 lines
766 B
Python
29 lines
766 B
Python
from transformers.models.mllama import configuration_mllama as mllama_hf_config
|
|
|
|
|
|
class MllamaTextConfig(mllama_hf_config.MllamaTextConfig):
|
|
'''
|
|
Use this class to override is_encoder_decoder:
|
|
- transformers regards mllama as is_encoder_decoder=False
|
|
- vllm needs is_encoder_decoder=True to enable cross-attention
|
|
'''
|
|
|
|
def __init__(
|
|
self,
|
|
**kwargs,
|
|
):
|
|
super().__init__(**kwargs)
|
|
self.is_encoder_decoder = True
|
|
|
|
|
|
class MllamaConfig(mllama_hf_config.MllamaConfig):
|
|
|
|
def __init__(
|
|
self,
|
|
text_config=None,
|
|
**kwargs,
|
|
):
|
|
if isinstance(text_config, dict):
|
|
text_config = MllamaTextConfig(**text_config)
|
|
super().__init__(text_config=text_config, **kwargs)
|