Merge remote-tracking branch 'origin/sfast_optimization'

.github/workflows/docker-create-ai-runner.yaml → .github/workflows/ai-runner-docker.yaml

@@ -1,7 +1,7 @@
 name: Build ai-runner base Docker image
 
 on:
-  check_run:
+  pull_request:
     paths:
       - "runner/**"
       - "!runner/.devcontainer/**"
@@ -13,7 +13,6 @@ on:
     paths:
       - "runner/**"
       - "!runner/.devcontainer/**"
-  workflow_dispatch:
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
@@ -22,7 +21,7 @@ concurrency:
 jobs:
   docker:
     name: Docker image generation
-    if: github.event_name == 'push' || github.event.pull_request.head.repo.full_name == github.repository || github.event_name == 'workflow_dispatch'
+    if: github.event_name == 'push' || github.event.pull_request.head.repo.full_name == github.repository
     permissions:
       packages: write
       contents: read
@@ -50,7 +49,7 @@ jobs:
         uses: docker/metadata-action@v5
         with:
           images: |
-            jjassonn69/ai-runner
+            livepeer/ai-runner
           tags: |
             type=sha
             type=ref,event=pr
@@ -75,5 +74,5 @@ jobs:
           tags: ${{ steps.meta.outputs.tags }}
           file: "Dockerfile"
           labels: ${{ steps.meta.outputs.labels }}
-          cache-from: type=registry,ref=jjassonn69/build:cache
-          cache-to: type=registry,ref=jjassonn69/build:cache,mode=max
+          cache-from: type=registry,ref=livepeerci/build:cache
+          cache-to: type=registry,ref=livepeerci/build:cache,mode=max

.github/workflows/trigger-upstream-openapi-sync.yaml

@@ -0,0 +1,30 @@
+name: Trigger upstream OpenAPI sync
+
+on:
+  push:
+    paths:
+      - "runner/openapi.json"
+  workflow_dispatch:
+
+jobs:
+  trigger-upstream-openapi-sync:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Trigger docs AI OpenAPI spec update
+        uses: peter-evans/repository-dispatch@v3
+        with:
+          token: ${{ secrets.DOCS_TRIGGER_PAT }}
+          repository: livepeer/docs
+          event-type: update-ai-openapi
+          client-payload: '{"sha": "${{ github.sha }}"}'
+
+      - name: Trigger SDK generation
+        uses: peter-evans/repository-dispatch@v3
+        with:
+          token: ${{ secrets.SDKS_TRIGGER_PAT }}
+          repository: livepeer/livepeer-ai-sdks
+          event-type: update-ai-openapi
+          client-payload: '{"sha": "${{ github.sha }}"}'

.github/workflows/validate-openapi-on-push.yaml → .github/workflows/validate-openapi-on-pr.yaml

@@ -1,9 +1,6 @@
 name: Check OpenAPI spec and Golang bindings
 
 on:
-  push:
-    branches:
-      - main
   pull_request:
 
 jobs:
@@ -46,4 +43,4 @@ jobs:
           if ! git diff --exit-code; then
             echo "::error::Go bindings have changed. Please run 'make' at the root of the repository and commit the changes."
             exit 1
-          fi
+          fi

runner/app/pipelines/optim/sfast.py

@@ -1,7 +1,7 @@
 """This module provides a function to enable StableFast optimization for the pipeline.
 
-For more information, see the DeepCache project on GitHub: https://github.com/chengzeyi/stable-fast
-"""  # noqa: E501
+For more information, see the Stable Fast project on GitHub: https://github.com/chengzeyi/stable-fast
+"""
 
 import logging
 

runner/app/pipelines/upscale.py

@@ -73,18 +73,18 @@ def __init__(self, model_id: str):
                 # TODO: Retrieve defaults from Pydantic class in route.
                 warmup_kwargs = {
                     "prompt": "Upscaling the pipeline with sfast enabled",
-                    "image": PIL.Image.new("RGB", (576, 1024)),
+                    "image": PIL.Image.new("RGB", (400, 400)), # anything higher than this size cause the model to OOM
                 }
 
-                logger.info("Warming up ImageToVideoPipeline pipeline...")
+                logger.info("Warming up Upscale pipeline...")
                 total_time = 0
                 for ii in range(SFAST_WARMUP_ITERATIONS):
                     t = time.time()
                     try:
                         self.ldm(**warmup_kwargs).images
                     except Exception as e:
                         # FIXME: When out of memory, pipeline is corrupted.
-                        logger.error(f"ImageToVideoPipeline warmup error: {e}")
+                        logger.error(f"Upscale pipeline warmup error: {e}")
                         raise e
                     iteration_time = time.time() - t
                     total_time += iteration_time