truss 0.11.8rc7__tar.gz → 0.11.8rc10__tar.gz

{truss-0.11.8rc7 → truss-0.11.8rc10}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: truss
-Version: 0.11.8rc7
+Version: 0.11.8rc10
 Summary: A seamless bridge from model development to model delivery
 Project-URL: Repository, https://github.com/basetenlabs/truss
 Project-URL: Homepage, https://truss.baseten.co

{truss-0.11.8rc7 → truss-0.11.8rc10}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "truss"
-version = "0.11.8rc7"
+version = "0.11.8rc10"
 description = "A seamless bridge from model development to model delivery"
 authors = [
     { name = "Pankaj Gupta", email = "no-reply@baseten.co" },

{truss-0.11.8rc7 → truss-0.11.8rc10}/truss/base/trt_llm_config.py

@@ -329,9 +329,16 @@ pip install truss==0.10.8
             raise ValueError("Using fp8 context fmha requires paged context fmha")
         if (
             self.plugin_configuration.use_fp8_context_fmha
-            and not self.quantization_type == TrussTRTLLMQuantizationType.FP8_KV
+            and self.quantization_type
+            not in (
+                TrussTRTLLMQuantizationType.FP8_KV,
+                TrussTRTLLMQuantizationType.FP4_KV,
+            )
         ):
-            raise ValueError("Using fp8 context fmha requires fp8 kv cache dtype")
+            raise ValueError(
+                "Using fp8 context fmha requires fp8 kv, or fp4 with kv cache dtype"
+            )
+
         return self
 
     def _validate_speculator_config(self):

{truss-0.11.8rc7 → truss-0.11.8rc10}/truss/templates/control/control/application.py

@@ -2,13 +2,12 @@ import asyncio
 import logging
 import logging.config
 import re
-import traceback
 from pathlib import Path
-from typing import Awaitable, Callable, Dict
+from typing import Dict
 
 import httpx
 from endpoints import control_app
-from fastapi import FastAPI, Request, Response
+from fastapi import FastAPI
 from fastapi.responses import JSONResponse
 from helpers.errors import ModelLoadFailed, PatchApplicatonError
 from helpers.inference_server_controller import InferenceServerController
@@ -17,45 +16,22 @@ from helpers.inference_server_starter import async_inference_server_startup_flow
 from helpers.truss_patch.model_container_patch_applier import ModelContainerPatchApplier
 from shared import log_config
 from starlette.datastructures import State
-from starlette.middleware.base import BaseHTTPMiddleware
-
-SANITIZED_EXCEPTION_FRAMES = 2
-
-
-class SanitizedExceptionMiddleware(BaseHTTPMiddleware):
-    def __init__(self, app, num_frames: int = SANITIZED_EXCEPTION_FRAMES):
-        super().__init__(app)
-        self.num_frames = num_frames
-
-    async def dispatch(
-        self, request: Request, call_next: Callable[[Request], Awaitable[Response]]
-    ) -> Response:
-        try:
-            return await call_next(request)
-        except Exception as exc:
-            sanitized_traceback = self._create_sanitized_traceback(exc)
-            if hasattr(request.app.state, "logger"):
-                request.app.state.logger.error(f"Error:\n{sanitized_traceback}")
-
-            if isinstance(exc, ModelLoadFailed):
-                return JSONResponse({"error": str(exc)}, status_code=503)
-            elif isinstance(exc, PatchApplicatonError):
-                error_type = _camel_to_snake_case(type(exc).__name__)
-                return JSONResponse(
-                    {"error": {"type": error_type, "msg": str(exc)}}, status_code=400
-                )
-            else:
-                return JSONResponse(
-                    {"error": {"type": "unknown", "msg": str(exc)}}, status_code=500
-                )
-
-    def _create_sanitized_traceback(self, error: Exception) -> str:
-        tb_lines = traceback.format_tb(error.__traceback__)
-        if tb_lines and self.num_frames > 0:
-            selected_frames = tb_lines[-self.num_frames :]
-            return "".join(selected_frames).rstrip()
-        else:
-            return f"{type(error).__name__}: {error}"
+
+
+async def handle_patch_error(_, exc):
+    error_type = _camel_to_snake_case(type(exc).__name__)
+    return JSONResponse(content={"error": {"type": error_type, "msg": str(exc)}})
+
+
+async def generic_error_handler(_, exc):
+    return JSONResponse(
+        content={"error": {"type": "unknown", "msg": f"{type(exc)}: {exc}"}}
+    )
+
+
+async def handle_model_load_failed(_, error):
+    # Model load failures should result in 503 status
+    return JSONResponse({"error": str(error)}, 503)
 
 
 def create_app(base_config: Dict):
@@ -106,10 +82,14 @@ def create_app(base_config: Dict):
     app = FastAPI(
         title="Truss Live Reload Server",
         on_startup=[start_background_inference_startup],
+        exception_handlers={
+            PatchApplicatonError: handle_patch_error,
+            ModelLoadFailed: handle_model_load_failed,
+            Exception: generic_error_handler,
+        },
     )
     app.state = app_state
     app.include_router(control_app)
-    app.add_middleware(SanitizedExceptionMiddleware)
 
     @app.on_event("shutdown")
     def on_shutdown():

{truss-0.11.8rc7 → truss-0.11.8rc10}/truss/templates/control/control/endpoints.py

@@ -5,7 +5,6 @@ from typing import Any, Callable, Dict, Optional, Protocol
 import httpx
 from fastapi import APIRouter, WebSocket
 from fastapi.responses import JSONResponse, StreamingResponse
-from helpers.errors import ModelLoadFailed, ModelNotReady
 from httpx_ws import AsyncWebSocketSession, WebSocketDisconnect, aconnect_ws
 from httpx_ws import _exceptions as httpx_ws_exceptions
 from starlette.requests import ClientDisconnect, Request
@@ -14,6 +13,11 @@ from starlette.websockets import WebSocketDisconnect as StartletteWebSocketDisco
 from tenacity import RetryCallState, Retrying, retry_if_exception_type, wait_fixed
 from wsproto.events import BytesMessage, TextMessage
 
+from truss.templates.control.control.helpers.errors import (
+    ModelLoadFailed,
+    ModelNotReady,
+)
+
 INFERENCE_SERVER_START_WAIT_SECS = 60
 BASE_RETRY_EXCEPTIONS = (
     retry_if_exception_type(httpx.ConnectError)

{truss-0.11.8rc7 → truss-0.11.8rc10}/truss/templates/docker_server/proxy.conf.jinja

@@ -35,6 +35,16 @@ server {
         proxy_pass http://127.0.0.1:{{server_port}};
     }
     # Predict
+    location ~ ^(/v1/models/model:predict|/v1/websocket)$ {
+        proxy_redirect off;
+        proxy_read_timeout 18030s;
+        proxy_http_version 1.1;
+
+        rewrite ^/v1/models/model:predict$ {{server_endpoint}} break;
+
+        proxy_pass http://127.0.0.1:{{server_port}};
+    }
+
     location ~ ^/v1/models/model:predict$ {
         proxy_redirect off;
         proxy_read_timeout 18030s;