nroggendorff commited on
Commit
5c02f21
·
verified ·
1 Parent(s): 059540c

Update Dockerfile

Browse files
Files changed (1) hide show
  1. Dockerfile +18 -1
Dockerfile CHANGED
@@ -10,4 +10,21 @@ ENV DATASETS_CACHE=$HOME/.cache/huggingface/datasets \
10
  ENV TOKENIZERS_PARALLELISM=false \
11
  PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:128 \
12
  FLASH_ATTENTION=1 \
13
- ACCELERATE_USE_DEEPSPEED=true
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
10
  ENV TOKENIZERS_PARALLELISM=false \
11
  PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:128 \
12
  FLASH_ATTENTION=1 \
13
+ ACCELERATE_USE_DEEPSPEED=true
14
+
15
+ ENV NCCL_TIMEOUT=1800000 \
16
+ NCCL_BLOCKING_WAIT=1 \
17
+ NCCL_ASYNC_ERROR_HANDLING=1 \
18
+ NCCL_DEBUG=WARN \
19
+ NCCL_P2P_DISABLE=1 \
20
+ NCCL_IB_DISABLE=1 \
21
+ NCCL_NET_GDR_LEVEL=0 \
22
+ NCCL_SOCKET_IFNAME=lo \
23
+ NCCL_TREE_THRESHOLD=0
24
+
25
+ ENV TORCH_NCCL_BLOCKING_WAIT=1 \
26
+ TORCH_NCCL_ASYNC_ERROR_HANDLING=1 \
27
+ TORCH_DISTRIBUTED_DEBUG=DETAIL
28
+
29
+ ENV CUDA_LAUNCH_BLOCKING=0 \
30
+ CUDNN_BENCHMARK=1