bump up version minor number

lib/metrics.py

@@ -146,6 +146,7 @@ class Metrics:
     def _set_mtoken(self, mtoken: str) -> None:
         self.mtoken = mtoken
 
+
     #######################################Private#######################################
 
     async def __send_delete_requests_and_reset(self):
@@ -216,7 +217,7 @@ class Metrics:
                 id=self.id,
                 mtoken=self.mtoken,
                 version=self.version,
-                loadtime=(loadtime_snapshot or 0.0), 
+                loadtime=(loadtime_snapshot or 0.0),
                 new_load=self.model_metrics.workload_processing,
                 cur_load=self.model_metrics.cur_load,
                 rej_load=self.model_metrics.workload_rejected,
@@ -280,7 +281,6 @@ class Metrics:
 
         if sent:
             # clear the one-shot loadtime only if we actually sent *this* value
-            self.system_metrics.reset(expected=loadtime_snapshot)
             self.update_pending = False
             self.model_metrics.reset()
             self.last_metric_update = time.time()

lib/server.py

@@ -3,58 +3,38 @@ import logging
 from typing import List
 import ssl
 from asyncio import run, gather
-import asyncio
+
 
 from lib.backend import Backend
-from lib.metrics import Metrics
 from aiohttp import web
 
 log = logging.getLogger(__file__)
 
 
 def start_server(backend: Backend, routes: List[web.RouteDef], **kwargs):
-    try:
+    log.debug("getting certificate...")
-        log.debug("getting certificate...")
+    use_ssl = os.environ.get("USE_SSL", "false") == "true"
-        use_ssl = os.environ.get("USE_SSL", "false") == "true"
+    if use_ssl is True:
-        if use_ssl is True:
+        ssl_context = ssl.create_default_context(ssl.Purpose.CLIENT_AUTH)
-            ssl_context = ssl.create_default_context(ssl.Purpose.CLIENT_AUTH)
+        ssl_context.load_cert_chain(
-            ssl_context.load_cert_chain(
+            certfile="/etc/instance.crt",
-                certfile="/etc/instance.crt",
+            keyfile="/etc/instance.key",
-                keyfile="/etc/instance.key",
+        )
-            )
+    else:
-        else:
+        ssl_context = None
-            ssl_context = None
 
-        async def main():
+    async def main():
-            log.debug("starting server...")
+        log.debug("starting server...")
-            app = web.Application()
+        app = web.Application()
-            app.add_routes(routes)
+        app.add_routes(routes)
-            runner = web.AppRunner(app)
+        runner = web.AppRunner(app)
-            await runner.setup()
+        await runner.setup()
-            site = web.TCPSite(
+        site = web.TCPSite(
-                runner,
+            runner,
-                ssl_context=ssl_context,
+            ssl_context=ssl_context,
-                port=int(os.environ["WORKER_PORT"]),
+            port=int(os.environ["WORKER_PORT"]),
-                **kwargs
+            **kwargs
-            )
+        )
-            await gather(site.start(), backend._start_tracking())
+        await gather(site.start(), backend._start_tracking())
 
-        run(main())
+    run(main())
-
-    except Exception as e:
-        err_msg = f"PyWorker failed to launch: {e}"
-        log.error(err_msg)
-
-        async def beacon():
-            metrics = Metrics()
-            metrics._set_version(getattr(backend, "version", "0"))
-            metrics._set_mtoken(getattr(backend, "mtoken", ""))
-            try:
-                while True:
-                    metrics._model_errored(err_msg)
-                    await metrics._Metrics__send_metrics_and_reset()
-                    await asyncio.sleep(10)
-            finally:
-                await metrics.aclose()
-
-        run(beacon())

start_server.sh

@@ -41,14 +41,6 @@ echo_var DEBUG_LOG
 echo_var PYWORKER_LOG
 echo_var MODEL_LOG
 
-# if instance is rebooted, we want to clear out the log file so pyworker doesn't read lines
-# from the run prior to reboot. past logs are saved in $MODEL_LOG.old for debugging only
-if [ -e "$MODEL_LOG" ]; then
-    echo "Rotating model log at $MODEL_LOG to $MODEL_LOG.old"
-    cat "$MODEL_LOG" >> "$MODEL_LOG.old" 
-    : > "$MODEL_LOG"
-fi
-
 # Populate /etc/environment with quoted values
 if ! grep -q "VAST" /etc/environment; then
     env -0 | grep -zEv "^(HOME=|SHLVL=)|CONDA" | while IFS= read -r -d '' line; do
@@ -132,43 +124,9 @@ cd "$SERVER_DIR"
 
 echo "launching PyWorker server"
 
-set +e
+# if instance is rebooted, we want to clear out the log file so pyworker doesn't read lines
-python3 -m "workers.$BACKEND.server" |& tee -a "$PYWORKER_LOG"
+# from the run prior to reboot. past logs are saved in $MODEL_LOG.old for debugging only
-PY_STATUS=${PIPESTATUS[0]}
+[ -e "$MODEL_LOG" ] && cat "$MODEL_LOG" >> "$MODEL_LOG.old" && : > "$MODEL_LOG"
-set -e
 
-if [ "${PY_STATUS}" -ne 0 ]; then
+(python3 -m "workers.$BACKEND.server" |& tee -a "$PYWORKER_LOG") &
-  echo "PyWorker exited with status ${PY_STATUS}; notifying autoscaler..."
+echo "launching PyWorker server done"
-  ERROR_MSG="PyWorker exited: code ${PY_STATUS}"
-  MTOKEN="${MASTER_TOKEN:-}"
-  VERSION="${PYWORKER_VERSION:-0}"
-
-  IFS=',' read -r -a REPORT_ADDRS <<< "${REPORT_ADDR}"
-  for addr in "${REPORT_ADDRS[@]}"; do
-    curl -sS -X POST -H 'Content-Type: application/json' \
-      -d "$(cat <<JSON
-{
-  "id": ${CONTAINER_ID:-0},
-  "mtoken": "${MTOKEN}",
-  "version": "${VERSION}",
-  "loadtime": 0,
-  "new_load": 0,
-  "cur_load": 0,
-  "rej_load": 0,
-  "max_perf": 0,
-  "cur_perf": 0,
-  "error_msg": "${ERROR_MSG}",
-  "num_requests_working": 0,
-  "num_requests_recieved": 0,
-  "additional_disk_usage": 0,
-  "working_request_idxs": [],
-  "cur_capacity": 0,
-  "max_capacity": 0,
-  "url": "${URL}"
-}
-JSON
-)" "${addr%/}/worker_status/" || true
-  done
-fi
-
-echo "launching PyWorker server done"

workers/openai/README.md

@@ -8,13 +8,14 @@ This is the base PyWorker for OpenAI compatible inference servers.  See the [Ser
 
 This worker is compatible with any backend API that properly implements the `/v1/completions` and `/v1/chat/completions` endpoints.  We currently have three templates you can choose from but you can also create your own without having to modify the PyWorker.
 
-- [vLLM](https://cloud.vast.ai/?ref_id=62897&creator_id=62897&name=vLLM%20(Serverless)) (recommended)
+- [vLLM](https://cloud.vast.ai/?ref_id=62897&creator_id=62897&name=vLLM%20%2B%20Qwen%2FQwen3-8B%20(Serverless)) (recommended)
 - [Ollama](https://cloud.vast.ai/?ref_id=62897&creator_id=62897&name=Ollama%20%2B%20Qwen3%3A32b%20(Serverless))
+- [HuggingFace TGI](https://cloud.vast.ai/?ref_id=62897&creator_id=62897&name=TGI%20%2B%20Qwen3-8B%20(Serverless))
 
 
 All of these templates can be configured via the template interface.  You may want to change the model or startup arguments, depending on the template you selected.
 
-2. Follow the [getting started guide](https://docs.vast.ai/documentation/serverless/quickstart) for help with configuring your serverless setup.  For testing, we recommend that you use the default options presented by the web interface.
+2. Follow the [getting started guide](https://docs.vast.ai/serverless/getting-started) for help with configuring your serverless setup.  For testing, we recommend that you use the default options presented by the web interface.
 
 ## Client Setup (Demo)
 
@@ -33,30 +34,12 @@ uv pip install -r requirements.txt
 
 Several examples have been provided in the client to help you get started with your own implementation.
 
-First, set your API key as an environment variable:
+### Completions
+
+Call to `/v1/completions` with json response
 
 ```bash
-export VAST_API_KEY=<your_api_key>
+python -m workers.openai.client -k <API_KEY> -e <ENDPOINT_NAME> --completion --model <MODEL_NAME>
-```
-
-The `--model` and `--endpoint` flags are optional. If not provided, they default to `Qwen/Qwen3-8B` and `my-vllm-endpoint` respectively.
-
-### Chat Completion (streaming)
-
-Call to `/v1/chat/completions` with streaming response
-
-```bash
-python -m workers.openai.client --chat-stream --endpoint <ENDPOINT_NAME> --model <MODEL_NAME>
-```
-
-### Interactive Chat (streaming)
-
-Interactive session with calls to `/v1/chat/completions`.
-
-Type `clear` to clear the chat history or `quit` to exit.
-
-```bash
-python -m workers.openai.client --interactive --endpoint <ENDPOINT_NAME> --model <MODEL_NAME>
 ```
 
 ### Chat Completion (json)
@@ -64,7 +47,15 @@ python -m workers.openai.client --interactive --endpoint <ENDPOINT_NAME> --model
 Call to `/v1/chat/completions` with json response
 
 ```bash
-python -m workers.openai.client --chat --endpoint <ENDPOINT_NAME> --model <MODEL_NAME>
+python -m workers.openai.client -k <API_KEY> -e <ENDPOINT_NAME> --chat --model <MODEL_NAME>
+```
+
+### Chat Completion (streaming)
+
+Call to `/v1/chat/completions` with streaming response
+
+```bash
+python -m workers.openai.client -k <API_KEY> -e <ENDPOINT_NAME> --chat-stream --model <MODEL_NAME>
 ```
 
 ### Tool Use (json)
@@ -74,14 +65,16 @@ Call to `/v1/chat/completions` with tool and json response.
 This test defines a simple tool which will list the contents of the local pyworker directory.  The output is then analysed by the model.
 
 ```bash
-python -m workers.openai.client --tools --endpoint <ENDPOINT_NAME> --model <MODEL_NAME>
+python -m workers.openai.client -k <API_KEY> -e <ENDPOINT_NAME> --tools --model <MODEL_NAME>
 ```
 
-### Completions
+### Interactive Chat (streaming)
 
-Call to `/v1/completions` with json response
+Interactive session with calls to `/v1/chat/completions`.
+
+Type `clear` to clear the chat history or `quit` to exit.
 
 ```bash
-python -m workers.openai.client --completion --endpoint <ENDPOINT_NAME> --model <MODEL_NAME>
+python -m workers.openai.client -k <API_KEY> -e <ENDPOINT_NAME> --interactive --model <MODEL_NAME>
 ```
 

workers/openai/client.py

@@ -18,7 +18,7 @@ logging.basicConfig(
 log = logging.getLogger(__file__)
 
 # ---------------------- Prompts ----------------------
-COMPLETIONS_PROMPT = "Zebras are primarily grazers and can subsist on lower-quality vegetation. They are preyed on mainly by"
+COMPLETIONS_PROMPT = "the capital of USA is"
 CHAT_PROMPT = "Think step by step: Tell me about the Python programming language."
 TOOLS_PROMPT = (
     "Can you list the files in the current working directory and tell me what you see? "
@@ -97,9 +97,9 @@ def _tool_state_to_message_tool_calls(state: Dict[int, Dict[str, Any]]) -> List[
 
 
 # ---- OpenAI-compatible calls (non-streaming) ----
-async def call_completions(client: Serverless, *, model: str, prompt: str, endpoint_name: str, **kwargs) -> Dict[str, Any]:
+async def call_completions(client: Serverless, *, model: str, prompt: str, **kwargs) -> Dict[str, Any]:
 
-    endpoint = await client.get_endpoint(name=endpoint_name)
+    endpoint = await client.get_endpoint(name=ENDPOINT_NAME)
 
     payload = {
         "input": {
@@ -113,9 +113,9 @@ async def call_completions(client: Serverless, *, model: str, prompt: str, endpo
     resp = await endpoint.request("/v1/completions", payload, cost=payload["input"]["max_tokens"])
     return resp["response"]
 
-async def call_chat_completions(client: Serverless, *, model: str, messages: List[Dict[str, Any]], endpoint_name: str, **kwargs) -> Dict[str, Any]:
+async def call_chat_completions(client: Serverless, *, model: str, messages: List[Dict[str, Any]], **kwargs) -> Dict[str, Any]:
 
-    endpoint = await client.get_endpoint(name=endpoint_name)
+    endpoint = await client.get_endpoint(name=ENDPOINT_NAME)
 
     payload = {
         "input": {
@@ -132,9 +132,9 @@ async def call_chat_completions(client: Serverless, *, model: str, messages: Lis
     return resp["response"]
 
 # ---- Streaming variants ----
-async def stream_completions(client: Serverless, *, model: str, prompt: str, endpoint_name: str, **kwargs):
+async def stream_completions(client: Serverless, *, model: str, prompt: str, **kwargs):
 
-    endpoint = await client.get_endpoint(name=endpoint_name)
+    endpoint = await client.get_endpoint(name=ENDPOINT_NAME)
 
     payload = {
         "input": {
@@ -150,9 +150,9 @@ async def stream_completions(client: Serverless, *, model: str, prompt: str, end
     resp = await endpoint.request("/v1/completions", payload, cost=payload["input"]["max_tokens"], stream=True)
     return resp["response"]  # async generator
 
-async def stream_chat_completions(client: Serverless, *, model: str, messages: List[Dict[str, Any]], endpoint_name: str, **kwargs):
+async def stream_chat_completions(client: Serverless, *, model: str, messages: List[Dict[str, Any]], **kwargs):
 
-    endpoint = await client.get_endpoint(name=endpoint_name)
+    endpoint = await client.get_endpoint(name=ENDPOINT_NAME)
 
     payload = {
         "input": {
@@ -174,10 +174,9 @@ async def stream_chat_completions(client: Serverless, *, model: str, messages: L
 class APIDemo:
     """Demo and testing functionality for the API client"""
 
-    def __init__(self, client: Serverless, model: str, endpoint_name: str, tool_manager: Optional[ToolManager] = None):
+    def __init__(self, client: Serverless, model: str, tool_manager: Optional[ToolManager] = None):
         self.client = client
         self.model = model
-        self.endpoint_name = endpoint_name
         self.tool_manager = tool_manager or ToolManager()
 
     # ----- Streaming handler -----
@@ -186,15 +185,10 @@ class APIDemo:
         reasoning_content = ""
         printed_reasoning = False
         printed_answer = False
-        finish_reason = None
 
         async for chunk in stream:
             choice = (chunk.get("choices") or [{}])[0]
             delta = choice.get("delta", {})
-            
-            # Track finish reason
-            if choice.get("finish_reason"):
-                finish_reason = choice.get("finish_reason")
 
             # reasoning tokens
             rc = delta.get("reasoning_content")
@@ -225,8 +219,6 @@ class APIDemo:
                 print(f"Reasoning tokens: {len(reasoning_content.split())}")
             if printed_answer:
                 print(f"Response tokens: {len(full_response.split())}")
-            if finish_reason:
-                print(f"Finish reason: {finish_reason}")
 
         return full_response
     
@@ -239,7 +231,6 @@ class APIDemo:
             client=self.client,
             model=self.model,
             prompt=COMPLETIONS_PROMPT,
-            endpoint_name=self.endpoint_name,
             max_tokens=MAX_TOKENS,
             temperature=DEFAULT_TEMPERATURE,
         )
@@ -258,7 +249,6 @@ class APIDemo:
                 client=self.client,
                 model=self.model, 
                 messages=messages,
-                endpoint_name=self.endpoint_name,
                 max_tokens=MAX_TOKENS,
                 temperature=DEFAULT_TEMPERATURE
             )
@@ -271,7 +261,6 @@ class APIDemo:
                 client=self.client,
                 model=self.model, 
                 messages=messages,
-                endpoint_name=self.endpoint_name,
                 max_tokens=MAX_TOKENS,
                 temperature=DEFAULT_TEMPERATURE
             )
@@ -298,7 +287,6 @@ class APIDemo:
                 client=self.client,
                 model=self.model,
                 messages=messages,
-                endpoint_name=self.endpoint_name,
                 tools=minimal_tool,
                 tool_choice="none",
                 max_tokens=10
@@ -324,7 +312,6 @@ class APIDemo:
             client=self.client,
             model=self.model,
             messages=messages,
-            endpoint_name=self.endpoint_name,
             tools=self.tool_manager.get_ls_tool_definition(),
             tool_choice="auto",
             max_tokens=MAX_TOKENS,
@@ -402,7 +389,6 @@ class APIDemo:
             client=self.client,
             model=self.model,
             messages=messages,
-            endpoint_name=self.endpoint_name,
             max_tokens=MAX_TOKENS,
             temperature=DEFAULT_TEMPERATURE,
         )
@@ -441,6 +427,7 @@ class APIDemo:
         print("=" * 60)
         print("INTERACTIVE STREAMING CHAT")
         print("=" * 60)
+        print(f"Using model: {self.model}")
         print("Type 'quit' to exit, 'clear' to clear history")
         print()
 
@@ -466,8 +453,7 @@ class APIDemo:
                 stream = await stream_chat_completions(
                     client=self.client,
                     model=self.model, 
-                    messages=messages,
+                    messages=messages, 
-                    endpoint_name=self.endpoint_name,
                     max_tokens=MAX_TOKENS, 
                     temperature=0.7
                 )
@@ -487,8 +473,8 @@ class APIDemo:
 # ---------------------- CLI ----------------------
 def build_arg_parser() -> argparse.ArgumentParser:
     p = argparse.ArgumentParser(description="Vast vLLM Demo (Serverless SDK)")
-    p.add_argument("--model", default=DEFAULT_MODEL, help=f"Model to use for requests (default: {DEFAULT_MODEL})")
+    p.add_argument("--model", required=True, help="Model to use for requests (required)")
-    p.add_argument("--endpoint", default=ENDPOINT_NAME, help=f"Vast endpoint name (default: {ENDPOINT_NAME})")
+    p.add_argument("--endpoint", default="my-vllm-endpoint", help="Vast endpoint name (default: my-vllm-endpoint)")
 
     modes = p.add_mutually_exclusive_group(required=False)
     modes.add_argument("--completion", action="store_true", help="Test completions endpoint")
@@ -516,14 +502,12 @@ async def main_async():
         print("Please specify exactly one test mode")
         sys.exit(1)
 
-    print("=" * 60)
     print(f"Using model: {args.model}")
-    print(f"Using endpoint: {args.endpoint}")
+    print("=" * 60)
-
 
     try:
         async with Serverless() as client:
-            demo = APIDemo(client, args.model, args.endpoint, ToolManager())
+            demo = APIDemo(client, args.model, ToolManager())
 
             if args.completion:
                 await demo.demo_completions()

workers/openai/server.py

@@ -11,7 +11,6 @@ MODEL_SERVER_START_LOG_MSG = [
     "llama runner started",  # Ollama
     '"message":"Connected","target":"text_generation_router"',  # TGI
     '"message":"Connected","target":"text_generation_router::server"',  # TGI
-    "main: model loaded" # llama.cpp
 ]
 
 MODEL_SERVER_ERROR_LOG_MSGS = [
@@ -35,7 +34,6 @@ backend = Backend(
     model_server_url=os.environ["MODEL_SERVER_URL"],
     model_log_file=os.environ["MODEL_LOG"],
     allow_parallel_requests=True,
-    max_wait_time=600.0,
     benchmark_handler=CompletionsHandler(benchmark_runs=3, benchmark_words=256),
     log_actions=[
         *[(LogAction.ModelLoaded, info_msg) for info_msg in MODEL_SERVER_START_LOG_MSG],

workers/tgi/README.md

@@ -1,103 +1,19 @@
-# HuggingFace TGI PyWorker
+This is the base PyWorker for TGI, designed to create PyWorkers that can utilize various LLMs. It offers two primary endpoints:
 
-This is the base PyWorker for HuggingFace Text Generation Inference (TGI) servers. See the [Serverless documentation](https://docs.vast.ai/serverless) for guides and how-to's.
+1. `generate`: Generates the LLM's response to a given prompt in a single request.
+2. `generate_stream`: Streams the LLM's response token by token.
 
-## Instance Setup
+Both endpoints use the following API payload format:
-
-1. Pick a template
-
-This worker is compatible with any TGI backend. We have a template you can use or you can create your own.
-
-- [HuggingFace TGI](https://cloud.vast.ai/?ref_id=62897&creator_id=62897&name=TGI%20(Serverless))
-
-The template can be configured via the template interface. You may want to change the model or startup arguments.
-
-2. Follow the [getting started guide](https://docs.vast.ai/documentation/serverless/quickstart) for help with configuring your serverless setup. For testing, we recommend that you use the default options presented by the web interface.
-
-## Client Setup (Demo)
-
-1. Clone the PyWorker repository to your local machine and install the necessary requirements for running the test client.
-
-```bash
-git clone https://github.com/vast-ai/pyworker
-cd pyworker
-pip install uv
-uv venv -p 3.12
-source .venv/bin/activate
-uv pip install -r requirements.txt
-```
-
-## Using the Test Client
-
-The test client demonstrates both streaming and non-streaming generation using TGI's native API.
-
-First, set your API key as an environment variable:
-
-```bash
-export VAST_API_KEY=<your_api_key>
-```
-
-The `--endpoint` flag is optional. If not provided, it defaults to `my-tgi-endpoint`.
-
-### Generate (Streaming)
-
-Call to `/generate_stream` with streaming response:
-
-```bash
-python -m workers.tgi.client --generate-stream --endpoint <ENDPOINT_NAME>
-```
-
-### Generate (Non-Streaming)
-
-Call to `/generate` with json response:
-
-```bash
-python -m workers.tgi.client --generate --endpoint <ENDPOINT_NAME>
-```
-
-### Interactive Session (Streaming)
-
-Interactive session with streaming responses. Type `quit` to exit.
-
-```bash
-python -m workers.tgi.client --interactive --endpoint <ENDPOINT_NAME>
-```
-
-## API Endpoints
-
-TGI provides two primary endpoints:
-
-### Generate (Non-Streaming)
-
-`/generate` - Returns the complete response in a single request.
 
 ```json
 {
-  "inputs": "Your prompt here",
+  "inputs": "PROMPT",
   "parameters": {
-    "max_new_tokens": 1024,
+    "max_new_tokens": 250
-    "temperature": 0.7,
-    "return_full_text": false
   }
 }
 ```
 
-### Generate Stream (Streaming)
+Note that the max_new_tokens parameter, rather than the prompt size, impacts performance. For example, if an
-
+instance is benchmarked to process 100 tokens per second, a request with max_new_tokens = 200 will take
-`/generate_stream` - Streams the response token by token.
+approximately 2 seconds to complete.
-
-```json
-{
-  "inputs": "Your prompt here",
-  "parameters": {
-    "max_new_tokens": 1024,
-    "temperature": 0.7,
-    "do_sample": true,
-    "return_full_text": false
-  }
-}
-```
-
-## Performance Notes
-
-The `max_new_tokens` parameter (not the prompt size) primarily impacts performance. For example, if an instance is benchmarked to process 100 tokens per second, a request with `max_new_tokens = 200` will take approximately 2 seconds to complete.

workers/tgi/client.py

@@ -1,222 +1,61 @@
-import logging
-import json
-import os
-import sys
-import argparse
-
 from vastai import Serverless
 import asyncio
 
-# ---------------------- Logging ----------------------
+ENDPOINT_NAME = "my-tgi-endpoint" # Change this to match your endpoint name
-logging.basicConfig(
-    level=logging.DEBUG,
-    format="%(asctime)s[%(levelname)-5s] %(message)s",
-    datefmt="%Y-%m-%d %H:%M:%S",
-)
-log = logging.getLogger(__file__)
-
-# ---------------------- Defaults ----------------------
-DEFAULT_PROMPT = "Think step by step: Tell me about the Python programming language."
-
-ENDPOINT_NAME = "TGI-Prod2"       # change this to your TGI endpoint name
 MAX_TOKENS = 1024
-DEFAULT_TEMPERATURE = 0.7
+PROMPT = "Think step by step: Tell me about the Python programming language."
 
-
+async def call_generate(client: Serverless) -> None:
-# ---------------------- API Calls ----------------------
+    endpoint = await client.get_endpoint(name=ENDPOINT_NAME)
-async def call_generate(client: Serverless, *, endpoint_name: str, prompt: str, **kwargs) -> dict:
-    """Non-streaming generation via /generate endpoint"""
-    endpoint = await client.get_endpoint(name=endpoint_name)
 
     payload = {
-        "inputs": prompt,
+        "inputs": PROMPT,
         "parameters": {
-            "max_new_tokens": kwargs.get("max_tokens", MAX_TOKENS),
+            "max_new_tokens": MAX_TOKENS,
-            "temperature": kwargs.get("temperature", DEFAULT_TEMPERATURE),
+            "temperature": 0.7,
-            "return_full_text": False,
+            "return_full_text": False
         }
     }
-    log.debug("POST /generate %s", json.dumps(payload)[:500])
+
-    resp = await endpoint.request("/generate", payload, cost=payload["parameters"]["max_new_tokens"])
+    resp = await endpoint.request("/generate", payload, cost=MAX_TOKENS)
-    return resp["response"]
+
+    print(resp["response"]["generated_text"])
 
 
-async def call_generate_stream(client: Serverless, *, endpoint_name: str, prompt: str, **kwargs):
+async def call_generate_stream(client: Serverless) -> None:
-    """Streaming generation via /generate_stream endpoint"""
+    endpoint = await client.get_endpoint(name=ENDPOINT_NAME)
-    endpoint = await client.get_endpoint(name=endpoint_name)
 
     payload = {
-        "inputs": prompt,
+        "inputs": PROMPT,
         "parameters": {
-            "max_new_tokens": kwargs.get("max_tokens", MAX_TOKENS),
+            "max_new_tokens": MAX_TOKENS,
-            "temperature": kwargs.get("temperature", DEFAULT_TEMPERATURE),
+            "temperature": 0.7,
             "do_sample": True,
             "return_full_text": False,
         }
     }
-    log.debug("STREAM /generate_stream %s", json.dumps(payload)[:500])
+
     resp = await endpoint.request(
         "/generate_stream",
         payload,
-        cost=payload["parameters"]["max_new_tokens"],
+        cost=MAX_TOKENS,
         stream=True,
     )
-    return resp["response"]  # async generator
+    stream = resp["response"]
 
+    printed_answer = False
+    async for event in stream:
+        tok = (event.get("token") or {}).get("text")
+        if tok:
+            if not printed_answer:
+                printed_answer = True
+                print("Answer:\n", end="", flush=True)
+            print(tok, end="", flush=True)
 
-# ---------------------- Demo Runner ----------------------
+async def main():
-class APIDemo:
+    async with Serverless() as client:
-    """Demo and testing functionality for the TGI API client"""
+        await call_generate(client)
-
+        await call_generate_stream(client)
-    def __init__(self, client: Serverless, endpoint_name: str):
-        self.client = client
-        self.endpoint_name = endpoint_name
-
-    async def handle_streaming_response(self, stream) -> str:
-        """Process streaming response and print tokens"""
-        full_response = ""
-        printed_answer = False
-
-        async for event in stream:
-            tok = (event.get("token") or {}).get("text")
-            if tok:
-                if not printed_answer:
-                    printed_answer = True
-                    print("\n💬 Response: ", end="", flush=True)
-                print(tok, end="", flush=True)
-                full_response += tok
-
-        print()  # newline
-        if printed_answer:
-            print(f"\nStreaming completed. Response tokens: {len(full_response.split())}")
-
-        return full_response
-
-    async def demo_generate(self) -> None:
-        """Demo non-streaming generation"""
-        print("=" * 60)
-        print("GENERATE DEMO (NON-STREAMING)")
-        print("=" * 60)
-
-        response = await call_generate(
-            client=self.client,
-            endpoint_name=self.endpoint_name,
-            prompt=DEFAULT_PROMPT,
-            max_tokens=MAX_TOKENS,
-            temperature=DEFAULT_TEMPERATURE,
-        )
-        
-        print(f"\n💬 Response: {response.get('generated_text', '')}")
-        print(f"\nFull Response:\n{json.dumps(response, indent=2)}")
-
-    async def demo_generate_stream(self) -> None:
-        """Demo streaming generation"""
-        print("=" * 60)
-        print("GENERATE DEMO (STREAMING)")
-        print("=" * 60)
-
-        stream = await call_generate_stream(
-            client=self.client,
-            endpoint_name=self.endpoint_name,
-            prompt=DEFAULT_PROMPT,
-            max_tokens=MAX_TOKENS,
-            temperature=DEFAULT_TEMPERATURE,
-        )
-
-        try:
-            await self.handle_streaming_response(stream)
-        except Exception as e:
-            log.error("\nError during streaming: %s", e, exc_info=True)
-
-    async def interactive_chat(self) -> None:
-        """Interactive session with streaming generation"""
-        print("=" * 60)
-        print("INTERACTIVE STREAMING SESSION")
-        print("=" * 60)
-        print(f"Using endpoint: {self.endpoint_name}")
-        print("Type 'quit' to exit")
-        print()
-
-        while True:
-            try:
-                user_input = input("You: ").strip()
-
-                if user_input.lower() == "quit":
-                    print("👋 Goodbye!")
-                    break
-                elif not user_input:
-                    continue
-
-                print("Assistant: ", end="", flush=True)
-                stream = await call_generate_stream(
-                    client=self.client,
-                    endpoint_name=self.endpoint_name,
-                    prompt=user_input,
-                    max_tokens=MAX_TOKENS,
-                    temperature=DEFAULT_TEMPERATURE,
-                )
-
-                full_response = ""
-                async for event in stream:
-                    tok = (event.get("token") or {}).get("text")
-                    if tok:
-                        print(tok, end="", flush=True)
-                        full_response += tok
-                print()  # newline
-
-            except KeyboardInterrupt:
-                print("\n👋 Session interrupted. Goodbye!")
-                break
-            except Exception as e:
-                log.error("\nError: %s", e)
-                continue
-
-
-# ---------------------- CLI ----------------------
-def build_arg_parser() -> argparse.ArgumentParser:
-    p = argparse.ArgumentParser(description="Vast TGI Demo (Serverless SDK)")
-    p.add_argument("--endpoint", default=ENDPOINT_NAME, help=f"Vast endpoint name (default: {ENDPOINT_NAME})")
-
-    modes = p.add_mutually_exclusive_group(required=False)
-    modes.add_argument("--generate", action="store_true", help="Test generate endpoint (non-streaming)")
-    modes.add_argument("--generate-stream", action="store_true", help="Test generate endpoint with streaming")
-    modes.add_argument("--interactive", action="store_true", help="Start interactive streaming session")
-    return p
-
-
-async def main_async():
-    args = build_arg_parser().parse_args()
-
-    selected = sum([args.generate, args.generate_stream, args.interactive])
-    if selected == 0:
-        print("Please specify exactly one test mode:")
-        print("  --generate        : Test generate endpoint (non-streaming)")
-        print("  --generate-stream : Test generate endpoint with streaming")
-        print("  --interactive     : Start interactive streaming session")
-        print(f"\nExample: python {os.path.basename(sys.argv[0])} --generate-stream --endpoint my-tgi-endpoint")
-        sys.exit(1)
-    elif selected > 1:
-        print("Please specify exactly one test mode")
-        sys.exit(1)
-
-    print("=" * 60)
-    print(f"Using endpoint: {args.endpoint}")
-
-    try:
-        async with Serverless() as client:
-            demo = APIDemo(client, args.endpoint)
-
-            if args.generate:
-                await demo.demo_generate()
-            elif args.generate_stream:
-                await demo.demo_generate_stream()
-            elif args.interactive:
-                await demo.interactive_chat()
-
-    except Exception as e:
-        log.error("Error during test: %s", e, exc_info=True)
-        sys.exit(1)
-
 
 if __name__ == "__main__":
-    asyncio.run(main_async())
+    asyncio.run(main())