test vastai_sdk test package

.devcontainer/devcontainer.json

@@ -1,37 +0,0 @@
-// .devcontainer/devcontainer.json
-// Dev container for the Vast.ai serverless Ollama template.
-// Includes Docker-in-Docker so you can build and test images from inside the container.
-{
-  "name": "vast.ai-serverless-ollama",
-  "image": "mcr.microsoft.com/devcontainers/base:trixie",
-  "features": {
-    "ghcr.io/devcontainers/features/python:1": {
-      "installTools": true,
-      "version": "3.12"
-    },
-    "ghcr.io/devcontainers/features/docker-in-docker:3.0.1": {
-      "moby": false,
-      "version": "latest",
-      "installDockerBuildx": true,
-      "dockerDashComposeVersion": "v2"
-    }
-  },
-  "runArgs": ["--privileged"],
-  "containerEnv": {
-    "DOCKER_BUILDKIT": "1"
-  },
-  "postCreateCommand": "python3 -m pip install --user --upgrade pip && python3 -m pip install --user -r requirements.txt pyyaml",
-  "customizations": {
-    "vscode": {
-      "extensions": [
-        "ms-python.python",
-        "ms-azuretools.vscode-docker"
-      ],
-      "settings": {
-        "python.defaultInterpreterPath": "/usr/bin/python3",
-        "terminal.integrated.defaultProfile.linux": "bash",
-        "docker.showStartPage": false
-      }
-    }
-  }
-}

README.md

@@ -9,7 +9,7 @@ This repository contains **example PyWorkers** used by Vast.ai’s default Serve
 - Optionally supports FIFO queueing when the backend cannot process concurrent requests
 - Detects readiness/failure from model logs and runs a benchmark to estimate throughput
 
-> Important: The **core PyWorker framework** (Worker, WorkerConfig, HandlerConfig, BenchmarkConfig, LogActionConfig) is provided by the **`vastai` / `vastai-sdk`** Python package (https://github.com/vast-ai/vast-sdk). This repo focuses on *worker implementations and examples*, not the framework internals.
+> Important: The **core PyWorker framework** (Worker, WorkerConfig, HandlerConfig, BenchmarkConfig, LogActionConfig) is provided by the **`vastai`** Python package (https://github.com/vast-ai/vast-cli). This repo focuses on *worker implementations and examples*, not the framework internals.
 
 ## Repository Purpose
 

lib/wheres-my-pyworker.txt

@@ -1,16 +1,16 @@
 # Where did the PyWorker code go?
-We have moved the PyWorker source code into the `vastai-sdk` Python SDK.
+We have moved the PyWorker source code into the `vastai` Python package.
 You can install it with
 ```
-pip install vastai-sdk
+pip install vastai
 ```
 
 All of the source code can be found here:
-https://github.com/vast-ai/vast-sdk
+https://github.com/vast-ai/vast-cli
 
 And can be imported from vastai.serverless.server.lib
 
-Serverless instances automatically run the start_server.sh script, which installs the vastai-sdk.
+Serverless instances automatically run the start_server.sh script, which installs the vastai package.
 This is how the PyWorker source code makes it onto your serverless instances.
 You provide a worker.py file in your PYWORKER_REPO, and the start_server.sh will 
 create and run a PyWorker according to your configuration defined in the file.

requirements.txt

@@ -1 +1 @@
-vastai-sdk
+vastai>=0.3.0

start_server.sh

@@ -53,7 +53,7 @@ JSON
     exit 1
 }
 
-function install_vastai_sdk() {
+function install_vastai() {
     local uv_flags=()
     if [ "${USE_SYSTEM_PYTHON:-}" = "true" ]; then
         uv_flags+=(--system --break-system-packages)
@@ -77,8 +77,8 @@ function install_vastai_sdk() {
 
     if [ -n "${SDK_VERSION:-}" ]; then
         echo "Installing vastai version ${SDK_VERSION}"
-        if ! uv pip install "${uv_flags[@]}" "vastai==${SDK_VERSION}"; then
+        if ! uv pip install --index-url https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple/ "${uv_flags[@]}" "vastai-sdk-vast==${SDK_VERSION}"; then
-            report_error_and_exit "Failed to install vastai==${SDK_VERSION}"
+            report_error_and_exit "Failed to install vastai-vast==${SDK_VERSION}"
         fi
         return 0
     fi
@@ -140,7 +140,7 @@ if [ "${USE_SYSTEM_PYTHON:-}" = "true" ]; then
             fi
         fi
     fi
-    install_vastai_sdk
+    install_vastai
     touch ~/.no_auto_tmux
 elif [ ! -d "$ENV_PATH" ]; then
     echo "setting up venv"
@@ -197,7 +197,7 @@ elif [ ! -d "$ENV_PATH" ]; then
         report_error_and_exit "Failed to install Python requirements"
     fi
 
-    install_vastai_sdk
+    install_vastai
 
     if ! touch ~/.no_auto_tmux; then
         report_error_and_exit "Failed to create ~/.no_auto_tmux"
@@ -237,7 +237,7 @@ else
             fi
         fi
 
-        install_vastai_sdk
+        install_vastai
     fi
 fi
 
@@ -318,8 +318,8 @@ if [ "$IS_DEPLOYMENT" = "true" ]; then
     # The s3_key exists in the DB as soon as the deployment is created, but the
     # actual upload may still be in flight from the client side.
 
-    # Install SDK (uses the install_vastai_sdk function which supports SDK_BRANCH/SDK_VERSION)
+    # Install SDK (uses the install_vastai function which supports SDK_BRANCH/SDK_VERSION)
-    install_vastai_sdk
+    install_vastai
     # Run deployment in serve mode
     export VAST_DEPLOYMENT_MODE=serve
     echo "Starting deployment: python3 $DEPLOY_DIR/deployment.py"
@@ -339,19 +339,19 @@ set +e
 PY_STATUS=1
 
 if [ -f "$SERVER_DIR/worker.py" ]; then
-    echo "Running worker.py"
+    echo "trying worker.py"
     python3 -m "worker" |& tee -a "$PYWORKER_LOG"
     PY_STATUS=${PIPESTATUS[0]}
 fi
 
 if [ "${PY_STATUS}" -ne 0 ] && [ -f "$SERVER_DIR/workers/$BACKEND/worker.py" ]; then
-    echo "Running workers.${BACKEND}.worker"
+    echo "trying workers.${BACKEND}.worker"
     python3 -m "workers.${BACKEND}.worker" |& tee -a "$PYWORKER_LOG"
     PY_STATUS=${PIPESTATUS[0]}
 fi
 
 if [ "${PY_STATUS}" -ne 0 ] && [ -f "$SERVER_DIR/workers/$BACKEND/server.py" ]; then
-    echo "Running workers.${BACKEND}.server"
+    echo "trying workers.${BACKEND}.server"
     python3 -m "workers.${BACKEND}.server" |& tee -a "$PYWORKER_LOG"
     PY_STATUS=${PIPESTATUS[0]}
 fi
@@ -365,4 +365,4 @@ if [ "${PY_STATUS}" -ne 0 ]; then
     report_error_and_exit "PyWorker exited with status ${PY_STATUS}"
 fi
 
-echo "PyWorker bootstrap complete"
+echo "launching PyWorker server done"

workers/openai/worker.py

@@ -1,19 +1,33 @@
+import nltk
 import random
 import os
-import re
-import logging
 
 from vastai import Worker, WorkerConfig, HandlerConfig, LogActionConfig, BenchmarkConfig
 
-logging.getLogger().setLevel(logging.WARNING)  # Only show warnings and errors
+# vLLM model configuration
-
-# Ollama model configuration
 MODEL_SERVER_URL           = 'http://127.0.0.1'
-MODEL_SERVER_PORT          = 11434
+MODEL_SERVER_PORT          = 18000
-MODEL_LOG_FILE             = '/var/log/onstart.log'
+MODEL_LOG_FILE             = '/var/log/portal/vllm.log'
-MODEL_HEALTHCHECK_ENDPOINT = "/"
+MODEL_HEALTHCHECK_ENDPOINT = "/health"
+
+# vLLM-specific log messages
+MODEL_LOAD_LOG_MSG = [
+    "Application startup complete.",
+]
+
+MODEL_ERROR_LOG_MSGS = [
+    "INFO exited: vllm",
+    "RuntimeError: Engine",
+    "Traceback (most recent call last):"
+]
+
+MODEL_INFO_LOG_MSGS = [
+    '"message":"Download'
+]
+
+nltk.download("words")
+WORD_LIST = nltk.corpus.words.words()
 
-# Ollama-specific log messages
 def request_parser(request):
     data = request
     if request.get("input") is not None:
@@ -22,24 +36,8 @@ def request_parser(request):
 
 
 def completions_benchmark_generator() -> dict:
-    # extract words from the python source code of the worker to create a list of words for generating prompts
-
-    WORD_LIST = []
-
-    # Try to load from perl copyright file first
-    try:
-        with open("/usr/share/doc/perl/copyright", 'r') as f:
-            source_code = f.read()
-            WORD_LIST = re.findall(r'\b\w+\b', source_code)
-    except (FileNotFoundError, IOError):
-        # Fallback to loading from python file
-        with open(__file__, 'r') as f:
-            source_code = f.read()
-            WORD_LIST = re.findall(r'\b\w+\b', source_code)
-
     prompt = " ".join(random.choices(WORD_LIST, k=int(250)))
     model = os.environ.get("MODEL_NAME")
-
     if not model:
         raise ValueError("MODEL_NAME environment variable not set")
 
@@ -79,9 +77,9 @@ worker_config = WorkerConfig(
         )
     ],
     log_action_config=LogActionConfig(
-        on_load=["llama_server: model loaded"],
+        on_load=MODEL_LOAD_LOG_MSG,
-        on_error=["Traceback (most recent call last):","Error:"],
+        on_error=MODEL_ERROR_LOG_MSGS,
-        #on_info=["load_tensors:","llama_context:","print_info:","llama_model_loader:"]
+        on_info=MODEL_INFO_LOG_MSGS
     )
 )
 

workers/tgi/worker.py

@@ -35,7 +35,7 @@ def benchmark_generator() -> dict:
     benchmark_data = {
         "inputs": prompt,
         "parameters": {
-            "max_new_tokens": 500,
+            "max_new_tokens": 128,
             "temperature": 0.7,
             "return_full_text": False
         }