requirements_gpu.txt

# =============================================================
# GPU 版本依赖 (适配 NVIDIA A100-SXM4-80GB x2, CUDA 12.8)
# 验证通过环境: torch 2.10.0+cu128, vllm 0.19.1, verl 0.7.1
# =============================================================

# --- Deep Learning Framework ---
torch==2.10.0
torchaudio==2.10.0
torchvision==0.25.0

# --- RL Training Framework ---
verl==0.7.1
megatron-core>=0.12.0

# --- Inference Engine ---
vllm==0.19.1

# --- Distributed Computing ---
ray>=2.47.0

# --- Attention (需从源码编译, 见下方说明) ---
# flash-attn 需要从源码构建以匹配 torch 版本:
#   pip install flash-attn --no-build-isolation --no-cache-dir
# 构建依赖: ninja, packaging, wheel
flash-attn>=2.7.0

# --- LLM / Agent ---
transformers>=4.43.0
sentence_transformers
openai

# --- Data & Math ---
numpy

# --- Cloud / External ---
vertexai

# --- Utilities ---
func_timeout
sql_metadata
firecrawl
hydra-core>=1.3.0
omegaconf>=2.3.0
fastapi
pydantic
sse_starlette
aiohttp
loguru
einops