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.
13 lines
419 B
Python
13 lines
419 B
Python
# Adapted from
|
|
# https://huggingface.co/nvidia/NVLM-D-72B/blob/main/configuration_nvlm_d.py
|
|
# --------------------------------------------------------
|
|
# NVLM-D
|
|
# Copyright (c) 2024 NVIDIA
|
|
# Licensed under Apache 2.0 License [see LICENSE for details]
|
|
# --------------------------------------------------------
|
|
from .internvl import InternVLChatConfig
|
|
|
|
|
|
class NVLM_D_Config(InternVLChatConfig):
|
|
model_type = 'NVLM_D'
|