Refactor of the Scheduler package to enable or setup for: distributed benchmarks, multi-turn requests, advanced stopping criteria, large concurrency, multi modality, evaluations

pyproject.toml

@@ -44,10 +44,13 @@ keywords = [
 ]
 dependencies = [
     "click>=8.0.0,<8.2.0",
+    "culsans~=0.9.0",
     "datasets",
+    "eval_type_backport",
     "ftfy>=6.0.0",
     "httpx[http2]<1.0.0",
     "loguru",
+    "msgpack",
     "numpy",
     "pillow",
     "protobuf",
@@ -139,6 +142,7 @@ ignore_missing_imports=true
 
 
 [tool.ruff]
+target-version = "py39"
 line-length = 88
 indent-width = 4
 exclude = ["build", "dist", "env", ".venv"]
@@ -149,15 +153,16 @@ indent-style = "space"
 
 [tool.ruff.lint]
 ignore = [
-    "PLR0913",
-    "TC001",
-    "COM812",
-    "ISC001",
-    "TC002",
+    "COM812",  # ignore trailing comma errors due to older Python versions
+    "PD011", # ignore .values usage since ruff assumes it's a Pandas DataFrame
+    "PLR0913",  # ignore too many arguments in function definitions
     "PLW1514", # allow Path.open without encoding
     "RET505", # allow `else` blocks
     "RET506", # allow `else` blocks
-    "PD011", # ignore .values usage since ruff assumes it's a Pandas DataFrame
+    "S311",  # allow standard pseudo-random generators
+    "TC001",  # ignore imports used only for type checking
+    "TC002",  # ignore imports used only for type checking
+    "TC003", # ignore imports used only for type checking
 ]
 select = [
     # Rules reference: https://docs.astral.sh/ruff/rules/

src/guidellm/backend/__init__.py

@@ -1,23 +1,24 @@
+"""
+Backend infrastructure for GuideLLM language model interactions.
+
+Provides abstract base classes, implemented backends, request/response objects,
+and timing utilities for standardized communication with LLM providers.
+"""
+
 from .backend import (
     Backend,
     BackendType,
 )
-from .openai import CHAT_COMPLETIONS_PATH, TEXT_COMPLETIONS_PATH, OpenAIHTTPBackend
-from .response import (
-    RequestArgs,
-    ResponseSummary,
-    StreamingResponseType,
-    StreamingTextResponse,
+from .objects import (
+    GenerationRequest,
+    GenerationRequestTimings,
+    GenerationResponse,
 )
 
 __all__ = [
-    "CHAT_COMPLETIONS_PATH",
-    "TEXT_COMPLETIONS_PATH",
     "Backend",
     "BackendType",
-    "OpenAIHTTPBackend",
-    "RequestArgs",
-    "ResponseSummary",
-    "StreamingResponseType",
-    "StreamingTextResponse",
+    "GenerationRequest",
+    "GenerationRequestTimings",
+    "GenerationResponse",
 ]