aiter attention-backend (default enabled on AMD/ROCm) (#6381)

.github/workflows/pr-test-amd.yml

@@ -44,7 +44,7 @@ jobs:
         run: bash scripts/amd_ci_install_dependency.sh
 
       - name: Evaluate Accuracy
-        timeout-minutes: 20
+        timeout-minutes: 30
         run: |
           bash scripts/amd_ci_exec.sh python3 test_eval_accuracy_large.py
           bash scripts/amd_ci_exec.sh python3 test_eval_fp8_accuracy.py
@@ -70,7 +70,7 @@ jobs:
         run: bash scripts/amd_ci_install_dependency.sh
 
       - name: Evaluate accuracy (TP=2)
-        timeout-minutes: 20
+        timeout-minutes: 30
         run: |
           bash scripts/amd_ci_exec.sh python3 test_moe_eval_accuracy_large.py
 
@@ -94,7 +94,7 @@ jobs:
         run: bash scripts/amd_ci_install_dependency.sh
 
       - name: MLA TEST
-        timeout-minutes: 20
+        timeout-minutes: 30
         run: |
           bash scripts/amd_ci_exec.sh python3 test_mla.py
 
@@ -118,28 +118,28 @@ jobs:
         run: bash scripts/amd_ci_install_dependency.sh
 
       - name: Benchmark single latency
-        timeout-minutes: 10
+        timeout-minutes: 20
         run: |
           bash scripts/amd_ci_exec.sh python3 -m unittest test_bench_one_batch.TestBenchOneBatch.test_bs1_small
           bash scripts/amd_ci_exec.sh python3 -m unittest test_bench_one_batch.TestBenchOneBatch.test_bs1_default
 
       - name: Benchmark online latency
-        timeout-minutes: 10
+        timeout-minutes: 15
         run: |
           bash scripts/amd_ci_exec.sh python3 -m unittest test_bench_serving.TestBenchServing.test_online_latency_default
 
       - name: Benchmark offline throughput
-        timeout-minutes: 10
+        timeout-minutes: 15
         run: |
           bash scripts/amd_ci_exec.sh python3 -m unittest test_bench_serving.TestBenchServing.test_offline_throughput_default
 
       - name: Benchmark offline throughput (Non-streaming, small batch size)
-        timeout-minutes: 10
+        timeout-minutes: 15
         run: |
           bash scripts/amd_ci_exec.sh python3 -m unittest test_bench_serving.TestBenchServing.test_offline_throughput_non_stream_small_batch_size
 
       - name: Benchmark online latency (EAGLE)
-        timeout-minutes: 10
+        timeout-minutes: 15
         run: |
           bash scripts/amd_ci_exec.sh python3 -m unittest test_bench_serving.TestBenchServing.test_online_latency_eagle
 
@@ -163,17 +163,17 @@ jobs:
         run: bash scripts/amd_ci_install_dependency.sh
 
       - name: Benchmark offline throughput (w/o RadixAttention)
-        timeout-minutes: 10
+        timeout-minutes: 15
         run: |
           bash scripts/amd_ci_exec.sh python3 -m unittest test_bench_serving.TestBenchServing.test_offline_throughput_without_radix_cache
 
       - name: Benchmark offline throughput (w/ Triton)
-        timeout-minutes: 10
+        timeout-minutes: 15
         run: |
           bash scripts/amd_ci_exec.sh python3 -m unittest test_bench_serving.TestBenchServing.test_offline_throughput_with_triton_attention_backend
 
       - name: Benchmark offline throughput (w/ FP8)
-        timeout-minutes: 10
+        timeout-minutes: 15
         run: |
           bash scripts/amd_ci_exec.sh python3 -m unittest test_bench_serving.TestBenchServing.test_offline_throughput_default_fp8
 
@@ -197,27 +197,27 @@ jobs:
         run: bash scripts/amd_ci_install_dependency.sh
 
       - name: Benchmark dummy grok (TP=2)
-        timeout-minutes: 20
+        timeout-minutes: 30
         run: |
           bash scripts/amd_ci_exec.sh python3 models/test_dummy_grok_models.py
 
       - name: Benchmark single latency (TP=2)
-        timeout-minutes: 20
+        timeout-minutes: 25
         run: |
           bash scripts/amd_ci_exec.sh python3 -m unittest test_bench_one_batch.TestBenchOneBatch.test_moe_tp2_bs1
 
       - name: Benchmark single latency + torch.compile (TP=2)
-        timeout-minutes: 20
+        timeout-minutes: 25
         run: |
           bash scripts/amd_ci_exec.sh python3 -m unittest test_bench_one_batch.TestBenchOneBatch.test_torch_compile_tp2_bs1
 
       - name: Benchmark offline throughput (TP=2)
-        timeout-minutes: 20
+        timeout-minutes: 25
         run: |
           bash scripts/amd_ci_exec.sh python3 -m unittest test_bench_serving.TestBenchServing.test_moe_offline_throughput_default
 
       - name: Benchmark offline throughput (w/o RadixAttention) (TP=2)
-        timeout-minutes: 20
+        timeout-minutes: 25
         run: |
           bash scripts/amd_ci_exec.sh python3 -m unittest test_bench_serving.TestBenchServing.test_moe_offline_throughput_without_radix_cache
 
@@ -241,7 +241,7 @@ jobs:
         run: bash scripts/amd_ci_install_dependency.sh
 
       - name: Run test
-        timeout-minutes: 30
+        timeout-minutes: 40
         run: |
           bash scripts/amd_ci_exec.sh python3 run_suite.py --suite per-commit-amd
 
@@ -265,7 +265,7 @@ jobs:
         run: bash scripts/amd_ci_install_dependency.sh
 
       - name: Run test
-        timeout-minutes: 30
+        timeout-minutes: 40
         run: |
           bash scripts/amd_ci_exec.sh python3 run_suite.py --suite per-commit-2-gpu-amd
 
@@ -289,7 +289,7 @@ jobs:
         run: bash scripts/amd_ci_install_dependency.sh
 
       - name: Run test
-        timeout-minutes: 30
+        timeout-minutes: 40
         run: |
           bash scripts/amd_ci_exec.sh python3 run_suite.py --suite per-commit-8-gpu-amd