use ray

huggingface · Sep 21, 2024 · 3f820de · 3f820de
1 parent 192adfe
commit 3f820de
Show file tree

Hide file tree

Showing 5 changed files with 4 additions and 8 deletions.
diff --git a/.github/workflows/test_cli_cuda_vllm.yaml b/.github/workflows/test_cli_cuda_vllm.yaml
@@ -47,7 +47,7 @@ jobs:
 
       - name: Run tests
         run: |
-          pytest -x -s -k "cli and cuda and vllm and not tp"
+          FORCE_SERIAL=1 pytest -x -s -k "cli and cuda and vllm and not tp"
 
   run_cli_cuda_vllm_multi_gpu_tests:
     if: ${{
@@ -76,4 +76,4 @@ jobs:
 
       - name: Run tests
         run: |
-          pytest -x -s -k "cli and cuda and vllm and tp"
+          FORCE_SERIAL=1 pytest -x -s -k "cli and cuda and vllm and tp"
diff --git a/tests/configs/_memory_utilization_.yaml b/tests/configs/_memory_utilization_.yaml
diff --git a/tests/configs/_tp_.yaml b/tests/configs/_tp_.yaml
@@ -2,7 +2,8 @@ backend:
   device_ids: 0,1
   engine_args:
     tensor_parallel_size: 2
-    distributed_executor_backend: mp
+    disable_custom_all_reduce: true
+    distributed_executor_backend: ray
 
 # hydra/cli specific settings
 hydra:

diff --git a/tests/configs/cuda_inference_vllm_bloom.yaml b/tests/configs/cuda_inference_vllm_bloom.yaml
@@ -3,7 +3,6 @@ defaults:
   - _base_ # inherits from base config
   - _cuda_ # inherits from cuda config
   - _inference_ # inherits from inference config
-  - _memory_utilization_ # inherits from memory utilization config
   - _serving_mode_ # inherits from serving_mode config
   - _no_weights_ # inherits from no weights config
   - _bloom_ # inherits from bloom config

diff --git a/tests/configs/cuda_inference_vllm_bloom_tp.yaml b/tests/configs/cuda_inference_vllm_bloom_tp.yaml
@@ -3,7 +3,6 @@ defaults:
   - _base_ # inherits from base config
   - _cuda_ # inherits from cuda config
   - _inference_ # inherits from inference config
-  - _memory_utilization_ # inherits from memory utilization config
   - _serving_mode_ # inherits from serving_mode config
   - _no_weights_ # inherits from no weights config
   - _bloom_ # inherits from bloom config