| 1 | #!/bin/bash |
| 2 | |
| 3 | # Constants |
| 4 | SHARED_STORAGE_DIR="local_storage" |
| 5 | PREFILL_OUTPUT="prefill_output.txt" |
| 6 | DECODE_OUTPUT="decode_output.txt" |
| 7 | SYNC_DECODE_RECOVERED_OUTPUT="sync_decode_recovered_output.txt" |
| 8 | ASYNC_DECODE_RECOVERED_OUTPUT="async_decode_recovered_output.txt" |
| 9 | |
| 10 | # Cleanup |
| 11 | rm -rf "$SHARED_STORAGE_DIR" |
| 12 | rm -f "$PREFILL_OUTPUT" "$DECODE_OUTPUT" "$SYNC_DECODE_RECOVERED_OUTPUT" "$ASYNC_DECODE_RECOVERED_OUTPUT" |
| 13 | |
| 14 | # Run inference examples |
| 15 | VLLM_ENABLE_V1_MULTIPROCESSING=0 CUDA_VISIBLE_DEVICES=0 python3 prefill_example.py |
| 16 | VLLM_ENABLE_V1_MULTIPROCESSING=0 CUDA_VISIBLE_DEVICES=0 python3 decode_example.py |
| 17 | VLLM_ENABLE_V1_MULTIPROCESSING=0 CUDA_VISIBLE_DEVICES=0 python3 decode_example.py --simulate-failure |
| 18 | VLLM_ENABLE_V1_MULTIPROCESSING=0 CUDA_VISIBLE_DEVICES=0 python3 decode_example.py --simulate-failure --async-load |
| 19 | |
| 20 | # Compare outputs |
| 21 | if ! cmp -s "$DECODE_OUTPUT" "$SYNC_DECODE_RECOVERED_OUTPUT"; then |
| 22 | echo "❌ Outputs differ: sync recovery failed." |
| 23 | diff -u "$DECODE_OUTPUT" "$SYNC_DECODE_RECOVERED_OUTPUT" |
| 24 | exit 1 |
| 25 | fi |
| 26 | |
| 27 | if ! cmp -s "$DECODE_OUTPUT" "$ASYNC_DECODE_RECOVERED_OUTPUT"; then |
| 28 | echo "❌ Outputs differ: async recovery failed." |
| 29 | diff -u "$DECODE_OUTPUT" "$ASYNC_DECODE_RECOVERED_OUTPUT" |
| 30 | exit 1 |
| 31 | fi |
| 32 | |
| 33 | echo "✅ Outputs match: recovery successful." |