|
|
@@ -37,7 +37,7 @@ jobs:
|
|
|
|
|
|
python -m petals.cli.run_server --converted_model_name_or_path $MODEL_NAME --block_indices 0:12 \
|
|
|
--new_swarm --identity tests/test.id --host_maddrs /ip4/127.0.0.1/tcp/31337 --throughput 1 \
|
|
|
- --torch_dtype float32 --compression NONE --attn_cache_tokens 2048 &> server1.log &
|
|
|
+ --torch_dtype float32 --compression NONE --attn_cache_tokens 2048 --adapters $ADAPTER_NAME &> server1.log &
|
|
|
SERVER1_PID=$!
|
|
|
|
|
|
sleep 5 # wait for the first server to initialize DHT
|
|
|
@@ -46,7 +46,7 @@ jobs:
|
|
|
# ^-- server 1 multiaddr is determined by --identity and --host_maddrs
|
|
|
|
|
|
python -m petals.cli.run_server --converted_model_name_or_path $MODEL_NAME --block_indices 12:22 \
|
|
|
- --initial_peers $INITIAL_PEERS --throughput 1 --torch_dtype float32 &> server2.log &
|
|
|
+ --initial_peers $INITIAL_PEERS --throughput 1 --torch_dtype float32 --adapters $ADAPTER_NAME &> server2.log &
|
|
|
SERVER2_PID=$!
|
|
|
|
|
|
sleep 10 # wait for initial servers to declare blocks, then let server decide which blocks to serve
|
|
|
@@ -59,14 +59,11 @@ jobs:
|
|
|
--initial_peers $INITIAL_PEERS --throughput 1 --torch_dtype float32 --tensor_parallel_devices cpu cpu &> server4.log &
|
|
|
SERVER4_PID=$!
|
|
|
|
|
|
- python -m petals.cli.run_server --converted_model_name_or_path $MODEL_NAME --block_indices 0:24 --new_swarm --identity tests/test.id --host_maddrs /ip4/127.0.0.1/tcp/31337 --throughput 1 --torch_dtype float32 --compression NONE --attn_cache_tokens 2048 --adapters $ADAPTER_NAME &> server5.log &
|
|
|
- SERVER5_PID=$!
|
|
|
-
|
|
|
tail -n 100 -f server*.log &
|
|
|
LOGGER_PID=$!
|
|
|
sleep 30 # wait for servers to download layers
|
|
|
|
|
|
- kill -0 $SERVER1_PID $SERVER2_PID $SERVER3_PID $SERVER4_PID $SERVER5_PID # ensure all servers survived init
|
|
|
+ kill -0 $SERVER1_PID $SERVER2_PID $SERVER3_PID $SERVER4_PID # ensure all servers survived init
|
|
|
|
|
|
pytest tests --durations=0 --durations-min=1.0 -v
|
|
|
|
artek0chumak