pin version

.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 

start_server.sh

@@ -60,32 +60,32 @@ function install_vastai_sdk() {
     fi
     if [ "$FORCE_UPDATE" = true ]; then
         uv_flags+=(--force-reinstall)
-        echo "Force reinstalling vastai"
+        echo "Force reinstalling vastai-sdk"
     fi
 
-    # If SDK_BRANCH is set, install vastai from the vast-cli repo at that branch/tag/commit.
+    # If SDK_BRANCH is set, install vastai-sdk from the vast-sdk repo at that branch/tag/commit.
     if [ -n "${SDK_BRANCH:-}" ]; then
         if [ -n "${SDK_VERSION:-}" ]; then
             echo "WARNING: Both SDK_BRANCH and SDK_VERSION are set; using SDK_BRANCH=${SDK_BRANCH}"
         fi
-        echo "Installing vastai from https://github.com/vast-ai/vast-cli/ @ ${SDK_BRANCH}"
+        echo "Installing vastai-sdk from https://github.com/vast-ai/vast-sdk/ @ ${SDK_BRANCH}"
-        if ! uv pip install "${uv_flags[@]}" "vastai @ git+https://github.com/vast-ai/vast-cli.git@${SDK_BRANCH}"; then
+        if ! uv pip install "${uv_flags[@]}" "vastai-sdk @ git+https://github.com/vast-ai/vast-sdk.git@${SDK_BRANCH}"; then
-            report_error_and_exit "Failed to install vastai from vast-ai/vast-cli@${SDK_BRANCH}"
+            report_error_and_exit "Failed to install vastai-sdk from vast-ai/vast-sdk@${SDK_BRANCH}"
         fi
         return 0
     fi
 
     if [ -n "${SDK_VERSION:-}" ]; then
-        echo "Installing vastai version ${SDK_VERSION}"
+        echo "Installing vastai-sdk version ${SDK_VERSION}"
-        if ! uv pip install "${uv_flags[@]}" "vastai==${SDK_VERSION}"; then
+        if ! uv pip install "${uv_flags[@]}" "vastai-sdk==${SDK_VERSION}"; then
-            report_error_and_exit "Failed to install vastai==${SDK_VERSION}"
+            report_error_and_exit "Failed to install vastai-sdk==${SDK_VERSION}"
         fi
         return 0
     fi
 
-    echo "Installing default vastai"
+    echo "Installing default vastai-sdk"
-    if ! uv pip install "${uv_flags[@]}" vastai; then
+    if ! uv pip install "${uv_flags[@]}" vastai-sdk; then
-        report_error_and_exit "Failed to install vastai"
+        report_error_and_exit "Failed to install vastai-sdk"
     fi
 }
 
@@ -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
         }