Change CI, fix vLLM on Pascal (#33)

* Use GH runner for rocm * Update repositories * Install custom triton with pascal support for vllm * Dedupe after each repo on rocm * Install triton before dependencies * Update .gitignore * Use subdirectory for triton installation * Replace nodm with ly, add power options to openbox * Use latest release instead of latest commit on default branch for some projects * Add update repositories script * Reenable rocm on selfhosted, but manual only * Change summary generation * Remove triton & npm cache * Increase cowspace
sasha0552 · Apr 25, 2024 · ee4d935 · ee4d935
1 parent f5eb989
commit ee4d935
Show file tree

Hide file tree

Showing 23 changed files with 203 additions and 235 deletions.
diff --git a/.ci/actions.py b/.ci/actions.py
@@ -23,9 +23,9 @@ def main():
   render_template(f"{i}/gh-build-iso.yml.jinja2", f"{o}/gh-build-iso-cuda.yml"      , platform="cuda", type="normal")
   render_template(f"{i}/gh-build-iso.yml.jinja2", f"{o}/gh-build-iso-cuda-empty.yml", platform="cuda", type="empty")
   render_template(f"{i}/gh-build-iso.yml.jinja2", f"{o}/gh-build-iso-cuda-gui.yml"  , platform="cuda", type="gui")
-  render_template(f"{i}/sh-build-iso.yml.jinja2", f"{o}/sh-build-iso-rocm.yml"      , platform="rocm", type="normal", delay=0)
+  render_template(f"{i}/sh-build-iso.yml.jinja2", f"{o}/sh-build-iso-rocm.yml"      , platform="rocm", type="normal")
   render_template(f"{i}/gh-build-iso.yml.jinja2", f"{o}/gh-build-iso-rocm-empty.yml", platform="rocm", type="empty")
-  render_template(f"{i}/sh-build-iso.yml.jinja2", f"{o}/sh-build-iso-rocm-gui.yml"  , platform="rocm", type="gui", delay=30)
+  render_template(f"{i}/sh-build-iso.yml.jinja2", f"{o}/sh-build-iso-rocm-gui.yml"  , platform="rocm", type="gui")
 
 if __name__ == "__main__":
   main()
diff --git a/.ci/options.ini b/.ci/options.ini
@@ -11,12 +11,20 @@ vllm = https://github.com/vllm-project/vllm.git
 
 [revisions]
 automatic = de267390905335d353deab3e9ac2031432f4d305
-axolotl = 68601ec6ad1cc0e8cb855376586e6eef6a8aa270
+axolotl = v0.4.0
 comfyui = 8dc19e40d129c8ee049be7be2657458509717ba5
-koboldcpp = 593f08bb78ab6035f91cc7aa753975254f30ed82
-llamacpp = 784e11dea1f5ce9638851b2b0dddb107e2a609c8
-sillytavern = 47b656260595967ad70b749845f7b5b33ed3631c
+koboldcpp = v1.63
+llamacpp = b2724
+sillytavern = 1.11.8
 sillytavern_extras = 1d82f3a8607319d1e09a2f4749a09c564c18c320
-text_generation_webui = ad122361ea1dbc4b419214891b83f6b9ac12f947
-vllm = a395a638c2f18d549e7d01655cf7a6dbee566f91
+text_generation_webui = snapshot-2024-04-21
+vllm = v0.4.1
+
+[strategy]
+axolotl = release
+koboldcpp = release
+llamacpp = release
+sillytavern = release
+text_generation_webui = release
+vllm = release
 
diff --git a/.ci/template/gh-build-iso.yml.jinja2 b/.ci/template/gh-build-iso.yml.jinja2
@@ -60,8 +60,14 @@ jobs:
           # Exit on error
           set -eu
 
-          # Print checksums to summary
-          sha256sum out/* > "$GITHUB_STEP_SUMMARY"
+          # Start code section
+          echo '`' > "$GITHUB_STEP_SUMMARY"
+
+          # Print checksums
+          sha256sum out/* | sed 's/out\///' >> "$GITHUB_STEP_SUMMARY"
+
+          # End code section
+          echo '`' >> "$GITHUB_STEP_SUMMARY"
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v4

diff --git a/.ci/template/sh-build-iso.yml.jinja2 b/.ci/template/sh-build-iso.yml.jinja2
@@ -1,9 +1,7 @@
 name: Build ISO on selfhosted runner ({{ "CUDA" if type == "cuda" else "ROCm" }}, {{ type }})
 
 on:
-  push:
-    branches:
-    - main
+  workflow_dispatch:
 
 jobs:
   vm-start:
@@ -30,33 +28,6 @@ jobs:
           # Fix permissions for private key
           chmod 600 /tmp/private.key
 
-      - name: Wait {{ delay }} seconds
-        run: |
-          sleep {{ delay }}
-
-      - name: Wait until Github Actions Runner VM is stopped
-        run: |
-          ok="0"
-
-          while true; do
-            status=$(ssh runner "virsh domstate GithubActions")
-
-            if [ "$status" = "shutoff" ]; then
-              if [ "$ok" = "1" ]; then
-                echo "VM is shut off. Exiting."
-                break
-              else
-                sleep $((10 + $RANDOM % 10))
-                ok="1"
-              fi
-            else
-              ok="0"
-            fi
-
-            echo "VM is not shut off yet. Waiting for 5 seconds..."
-            sleep 5
-          done
-
       - name: Start Github Actions Runner VM
         run: |
           ssh runner "virsh start GithubActions"
@@ -109,8 +80,14 @@ jobs:
           # Exit on error
           set -eu
 
-          # Print checksums to summary
-          sha256sum out/* > "$GITHUB_STEP_SUMMARY"
+          # Start code section
+          echo '`' > "$GITHUB_STEP_SUMMARY"
+
+          # Print checksums
+          sha256sum out/* | sed 's/out\///' >> "$GITHUB_STEP_SUMMARY"
+
+          # End code section
+          echo '`' >> "$GITHUB_STEP_SUMMARY"
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v4
@@ -150,3 +127,6 @@ jobs:
 
           # Wait 2 minutes
           sleep 120
+
+          # Forcefully stop VM if needed
+          ssh runner "virsh destroy GithubActions" || true
diff --git a/.ci/update.py b/.ci/update.py
@@ -3,37 +3,68 @@
 import configparser
 import requests
 
-def fetch_latest_revision(url):
+def fetch_latest_revision(url, strategy):
   # github
   if url.startswith("https://github.com/") and url.endswith(".git"):
     # extract repo path
     path = url[19:-4]
 
-    # get commits from api
-    response = requests.get(f"https://api.github.com/repos/{path}/commits")
+    match strategy:
+      case "commit":
+        # get commits from api
+        response = requests.get(f"https://api.github.com/repos/{path}/commits")
 
-    # throw error if not success
-    response.raise_for_status()
+        # throw error if not success
+        response.raise_for_status()
 
-    # parse json
-    data = response.json()
+        # parse json
+        data = response.json()
 
-    # return first commit id
-    return data[0]["sha"]
+        # return first commit id
+        return data[0]["sha"]
 
-  raise ValueError(f"Unsupported: {url}")
+      case "release":
+        # get releases from api
+        response = requests.get(f"https://api.github.com/repos/{path}/releases")
+
+        # throw error if not success
+        response.raise_for_status()
+
+        # parse json
+        data = response.json()
+
+        # find first release
+        for release in data:
+          if not release["prerelease"]:
+            return release["tag_name"]
+
+  raise ValueError(f"Unsupported url or strategy ({url}, {strategy})")
 
 def main():
   # load options
   config = configparser.ConfigParser()
   config.read(".ci/options.ini")
 
   # default option value
-  config["revisions"] = {}
+  if "revisions" not in config:
+    config["revisions"] = {}
+
+  # default option value
+  if "strategy" not in config:
+    config["strategy"] = {}
 
   # iterate over repositories
   for key, value in config.items("repositories"):
-    config["revisions"][key] = fetch_latest_revision(value)
+    # update strategy
+    strategy = "commit"
+
+    # extract strategy
+    if key in config["strategy"]:
+      strategy = config["strategy"][key]
+
+    # if not locked, update
+    if strategy != "locked":
+      config["revisions"][key] = fetch_latest_revision(value, strategy)
 
   # write config back
   with open(".ci/options.ini", "w") as file:

diff --git a/.github/workflows/gh-build-iso-cuda-empty.yml b/.github/workflows/gh-build-iso-cuda-empty.yml
@@ -51,8 +51,14 @@ jobs:
           # Exit on error
           set -eu
 
-          # Print checksums to summary
-          sha256sum out/* > "$GITHUB_STEP_SUMMARY"
+          # Start code section
+          echo '`' > "$GITHUB_STEP_SUMMARY"
+
+          # Print checksums
+          sha256sum out/* | sed 's/out\///' >> "$GITHUB_STEP_SUMMARY"
+
+          # End code section
+          echo '`' >> "$GITHUB_STEP_SUMMARY"
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v4

diff --git a/.github/workflows/gh-build-iso-cuda-gui.yml b/.github/workflows/gh-build-iso-cuda-gui.yml
@@ -53,8 +53,14 @@ jobs:
           # Exit on error
           set -eu
 
-          # Print checksums to summary
-          sha256sum out/* > "$GITHUB_STEP_SUMMARY"
+          # Start code section
+          echo '`' > "$GITHUB_STEP_SUMMARY"
+
+          # Print checksums
+          sha256sum out/* | sed 's/out\///' >> "$GITHUB_STEP_SUMMARY"
+
+          # End code section
+          echo '`' >> "$GITHUB_STEP_SUMMARY"
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v4

diff --git a/.github/workflows/gh-build-iso-cuda.yml b/.github/workflows/gh-build-iso-cuda.yml
@@ -53,8 +53,14 @@ jobs:
           # Exit on error
           set -eu
 
-          # Print checksums to summary
-          sha256sum out/* > "$GITHUB_STEP_SUMMARY"
+          # Start code section
+          echo '`' > "$GITHUB_STEP_SUMMARY"
+
+          # Print checksums
+          sha256sum out/* | sed 's/out\///' >> "$GITHUB_STEP_SUMMARY"
+
+          # End code section
+          echo '`' >> "$GITHUB_STEP_SUMMARY"
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v4

diff --git a/.github/workflows/gh-build-iso-rocm-empty.yml b/.github/workflows/gh-build-iso-rocm-empty.yml
@@ -55,8 +55,14 @@ jobs:
           # Exit on error
           set -eu
 
-          # Print checksums to summary
-          sha256sum out/* > "$GITHUB_STEP_SUMMARY"
+          # Start code section
+          echo '`' > "$GITHUB_STEP_SUMMARY"
+
+          # Print checksums
+          sha256sum out/* | sed 's/out\///' >> "$GITHUB_STEP_SUMMARY"
+
+          # End code section
+          echo '`' >> "$GITHUB_STEP_SUMMARY"
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v4

diff --git a/.github/workflows/gh-update-repositories.yml b/.github/workflows/gh-update-repositories.yml
@@ -0,0 +1,25 @@
+name: Update repositories
+
+on:
+  schedule:
+    - cron: "0 0 * * 0"
+
+  workflow_dispatch:
+
+jobs:
+  update:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Update repositories
+        run: .ci/update.py
+
+      - name: Create pull request
+        uses: peter-evans/create-pull-request@v6
+        with:
+          branch: update-repositories/patch
+          commit-message: Update repositories
+          title: Update repositories
diff --git a/.github/workflows/sh-build-iso-rocm-gui.yml b/.github/workflows/sh-build-iso-rocm-gui.yml
@@ -1,9 +1,7 @@
 name: Build ISO on selfhosted runner (ROCm, gui)
 
 on:
-  push:
-    branches:
-    - main
+  workflow_dispatch:
 
 jobs:
   vm-start:
@@ -29,33 +27,6 @@ jobs:
           # Fix permissions for private key
           chmod 600 /tmp/private.key
 
-      - name: Wait 30 seconds
-        run: |
-          sleep 30
-
-      - name: Wait until Github Actions Runner VM is stopped
-        run: |
-          ok="0"
-
-          while true; do
-            status=$(ssh runner "virsh domstate GithubActions")
-
-            if [ "$status" = "shutoff" ]; then
-              if [ "$ok" = "1" ]; then
-                echo "VM is shut off. Exiting."
-                break
-              else
-                sleep $((10 + $RANDOM % 10))
-                ok="1"
-              fi
-            else
-              ok="0"
-            fi
-
-            echo "VM is not shut off yet. Waiting for 5 seconds..."
-            sleep 5
-          done
-
       - name: Start Github Actions Runner VM
         run: |
           ssh runner "virsh start GithubActions"
@@ -103,8 +74,14 @@ jobs:
           # Exit on error
           set -eu
 
-          # Print checksums to summary
-          sha256sum out/* > "$GITHUB_STEP_SUMMARY"
+          # Start code section
+          echo '`' > "$GITHUB_STEP_SUMMARY"
+
+          # Print checksums
+          sha256sum out/* | sed 's/out\///' >> "$GITHUB_STEP_SUMMARY"
+
+          # End code section
+          echo '`' >> "$GITHUB_STEP_SUMMARY"
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v4
@@ -142,4 +119,7 @@ jobs:
           ssh runner "virsh shutdown GithubActions"
 
           # Wait 2 minutes
-          sleep 120
+          sleep 120
+
+          # Forcefully stop VM if needed
+          ssh runner "virsh destroy GithubActions" || true