truss 0.11.8rc7__tar.gz → 0.11.8rc8__tar.gz

{truss-0.11.8rc7 → truss-0.11.8rc8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: truss
-Version: 0.11.8rc7
+Version: 0.11.8rc8
 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-0.11.8rc8}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "truss"
-version = "0.11.8rc7"
+version = "0.11.8rc8"
 description = "A seamless bridge from model development to model delivery"
 authors = [
     { name = "Pankaj Gupta", email = "no-reply@baseten.co" },

{truss-0.11.8rc7 → truss-0.11.8rc8}/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-0.11.8rc8}/truss/templates/control/control/application.py

@@ -10,15 +10,27 @@ import httpx
 from endpoints import control_app
 from fastapi import FastAPI, Request, Response
 from fastapi.responses import JSONResponse
-from helpers.errors import ModelLoadFailed, PatchApplicatonError
-from helpers.inference_server_controller import InferenceServerController
-from helpers.inference_server_process_controller import InferenceServerProcessController
-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
 
+from truss.templates.control.control.helpers.errors import (
+    ModelLoadFailed,
+    PatchApplicatonError,
+)
+from truss.templates.control.control.helpers.inference_server_controller import (
+    InferenceServerController,
+)
+from truss.templates.control.control.helpers.inference_server_process_controller import (
+    InferenceServerProcessController,
+)
+from truss.templates.control.control.helpers.inference_server_starter import (
+    async_inference_server_startup_flow,
+)
+from truss.templates.control.control.helpers.truss_patch.model_container_patch_applier import (
+    ModelContainerPatchApplier,
+)
+
 SANITIZED_EXCEPTION_FRAMES = 2
 
 
@@ -34,8 +46,7 @@ class SanitizedExceptionMiddleware(BaseHTTPMiddleware):
             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}")
+            request.app.state.logger.error(sanitized_traceback)
 
             if isinstance(exc, ModelLoadFailed):
                 return JSONResponse({"error": str(exc)}, status_code=503)

{truss-0.11.8rc7 → truss-0.11.8rc8}/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)