Update to vastai 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
-      }
-    }
-  }
-}

requirements.txt

@@ -1 +1,2 @@
-vastai-sdk
+vastai-sdk>=0.3.0
+nltk==3.9.4 

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
-
-# Ollama model configuration
+# vLLM model configuration
 MODEL_SERVER_URL           = 'http://127.0.0.1'
-MODEL_SERVER_PORT          = 11434
-MODEL_LOG_FILE             = '/var/log/onstart.log'
-MODEL_HEALTHCHECK_ENDPOINT = "/"
+MODEL_SERVER_PORT          = 18000
+MODEL_LOG_FILE             = '/var/log/portal/vllm.log'
+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_error=["Traceback (most recent call last):","Error:"],
-        #on_info=["load_tensors:","llama_context:","print_info:","llama_model_loader:"]
+        on_load=MODEL_LOAD_LOG_MSG,
+        on_error=MODEL_ERROR_LOG_MSGS,
+        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
         }