remove fastertransformer build and release in DJLServing (#1241)

deepjavalibrary · Oct 31, 2023 · 0995b65 · 0995b65
1 parent 6d982f4
commit 0995b65
Show file tree

Hide file tree

Showing 20 changed files with 14 additions and 1,024 deletions.
diff --git a/.github/workflows/docker-nightly-publish.yml b/.github/workflows/docker-nightly-publish.yml
@@ -15,7 +15,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        arch: [ cpu, cpu-full, pytorch-inf2, pytorch-cu118, fastertransformer, tensorrt-llm ]
+        arch: [ cpu, cpu-full, pytorch-inf2, pytorch-cu118, tensorrt-llm ]
     steps:
       - name: Clean disk space
         run: |

diff --git a/.github/workflows/llm_integration.yml b/.github/workflows/llm_integration.yml
diff --git a/.github/workflows/sagemaker-integration.yml b/.github/workflows/sagemaker-integration.yml
@@ -98,13 +98,6 @@ jobs:
           python3 llm/sagemaker-endpoint-tests.py opt-1-3-b djl ${{ github.event.inputs.mode || 'nightly' }}
           echo "sleep 30 seconds to allow endpoint deletion"
           sleep 30
-      - name: Test flan-t5-xxl
-        if: success() || failure()
-        working-directory: tests/integration
-        run: |
-          python3 llm/sagemaker-endpoint-tests.py flan-t5-xxl djl ${{ github.event.inputs.mode || 'nightly' }}
-          echo "sleep 30 seconds to allow endpoint deletion"
-          sleep 30
         
   endpoint-tests-group-2:
     runs-on: [ self-hosted, cpu ]
@@ -142,13 +135,6 @@ jobs:
           python3 llm/sagemaker-endpoint-tests.py gpt-neo-2-7-b no_code ${{ github.event.inputs.mode || 'nightly' }}
           echo "sleep 30 seconds to allow endpoint deletion"
           sleep 30
-      - name: Test bloom-7b1 no code FasterTransformer
-        if: success() || failure()
-        working-directory: tests/integration
-        run: |
-          python3 llm/sagemaker-endpoint-tests.py bloom-7b1 no_code ${{ github.event.inputs.mode || 'nightly' }}
-          echo "sleep 30 seconds to allow endpoint deletion"
-          sleep 30
       - name: Test DeepSpeed pythia-12b
         if: success() || failure()
         working-directory: tests/integration

diff --git a/engines/python/setup/djl_python/fastertransformer.py b/engines/python/setup/djl_python/fastertransformer.py
diff --git a/engines/python/src/main/java/ai/djl/python/engine/MpiEngineProvider.java b/engines/python/src/main/java/ai/djl/python/engine/MpiEngineProvider.java
@@ -34,16 +34,6 @@ public int getEngineRank() {
         return PyEngine.RANK + 1;
     }
 
-    /** {@code FtEngineProvider} is the alias of {@link MpiEngineProvider}. */
-    public static final class FtEngineProvider extends MpiEngineProvider {
-
-        /** {@inheritDoc} */
-        @Override
-        public String getEngineName() {
-            return "FasterTransformer";
-        }
-    }
-
     /** {@code DsEngineProvider} is the alias of {@link MpiEngineProvider}. */
     public static final class DsEngineProvider extends MpiEngineProvider {
 

diff --git a/engines/python/src/main/java/ai/djl/python/engine/PyModel.java b/engines/python/src/main/java/ai/djl/python/engine/PyModel.java
@@ -160,8 +160,6 @@ public void load(Path modelPath, String prefix, Map<String, ?> options) throws I
                     entryPoint = modelFile.toFile().getName();
                 } else if ("DeepSpeed".equals(engineName)) {
                     entryPoint = "djl_python.deepspeed";
-                } else if ("FasterTransformer".equals(engineName)) {
-                    entryPoint = "djl_python.fastertransformer";
                 } else if ("nc".equals(manager.getDevice().getDeviceType())
                         && pyEnv.getTensorParallelDegree() > 0) {
                     entryPoint = "djl_python.transformers_neuronx";

diff --git a/engines/python/src/main/resources/META-INF/services/ai.djl.engine.EngineProvider b/engines/python/src/main/resources/META-INF/services/ai.djl.engine.EngineProvider
@@ -1,4 +1,3 @@
 ai.djl.python.engine.PyEngineProvider
 ai.djl.python.engine.MpiEngineProvider
-ai.djl.python.engine.MpiEngineProvider$FtEngineProvider
 ai.djl.python.engine.MpiEngineProvider$DsEngineProvider
diff --git a/serving/docker/docker-compose.yml b/serving/docker/docker-compose.yml
@@ -28,11 +28,6 @@ services:
       target: base
       dockerfile: pytorch-cu118.Dockerfile
     image: "deepjavalibrary/djl-serving:${RELEASE_VERSION}pytorch-cu118${NIGHTLY}"
-  fastertransformer:
-    build:
-      context: .
-      dockerfile: fastertransformer.Dockerfile
-    image: "deepjavalibrary/djl-serving:${RELEASE_VERSION}fastertransformer${NIGHTLY}"
   tensorrt-llm:
     build:
       context: .