feat: add Chinese UI, Responses models, and mirrored Windows setup
Localize the web console and persist custom model and capability settings. Use OpenAI Responses exclusively and move runtime limits into environment configuration. Bootstrap uv and the virtual environment on Windows, support optional China mirrors, and package source distributions without environments, build output, or credentials. Update regression tests and repository quality checks for the new behavior.
This commit is contained in:
chenmingxuan committed
1 parent
371aa6df56
commit
2ddec53954
183 files changed
+3427
-11857
No files matched your search
@@ -14,9 +14,8 @@
|
||||
|
||||
"""VideoAnalyzer facade: entry point and shared runtime state.
|
||||
|
||||
The execution engines live in sibling modules (``chunk_native``,
|
||||
``audio_native``, ``universal_engine``, ``native_coordinator``,
|
||||
``gemini_files``, ``universal_tools``) and resolve their patchable
|
||||
The execution engines live in sibling modules (``chunk_runner``,
|
||||
``audio_runner``, ``universal_engine``, ``universal_tools``) and resolve their patchable
|
||||
collaborators through THIS module's namespace, so existing
|
||||
``mock.patch("artemis.agents.video_analyzer.video_analyzer.<name>")``
|
||||
targets keep working. Keep the imports below in place even when the facade
|
||||
@@ -36,8 +35,6 @@ import tempfile
|
||||
import time
|
||||
from typing import Any
|
||||
|
||||
from google import genai
|
||||
from google.genai import types
|
||||
from langchain_core.messages import (
|
||||
BaseMessage,
|
||||
HumanMessage,
|
||||
@@ -49,9 +46,6 @@ from artemis.agents.video_analyzer.conflict_resolution import (
|
||||
ConflictResolutionService,
|
||||
)
|
||||
from artemis.agents.video_analyzer.blackboard import get_video_blackboard
|
||||
from artemis.agents.video_analyzer.gemini_files import (
|
||||
cleanup_abandoned_gemini_files,
|
||||
)
|
||||
from artemis.agents.video_analyzer.reliability import (
|
||||
AgenticVideoDegraded,
|
||||
VideoCircuitBreaker,
|
||||
@@ -60,19 +54,11 @@ from artemis.agents.video_analyzer.reliability import (
|
||||
from artemis.agents.video_analyzer.universal_tools import (
|
||||
UNIVERSAL_MAIN_TOOLS,
|
||||
UNIVERSAL_SUBMIT_ANSWER_TOOL,
|
||||
build_native_tools_declaration,
|
||||
build_submit_answer_declaration,
|
||||
)
|
||||
from artemis.config import get_temp_dir, settings
|
||||
from artemis.constants import SAFETY_SETTINGS_BLOCK_NONE
|
||||
from artemis.context import ArtemisContext
|
||||
from artemis.controllers.controller_factory import get_controller
|
||||
from artemis.data_engine.trace import CURRENT_TRACE_ID, TraceSpan, trace
|
||||
from artemis.llm.google import (
|
||||
is_gemini_model,
|
||||
resolve_video_processing,
|
||||
strip_provider_prefix,
|
||||
)
|
||||
from artemis.llm.reliability import retry_policy_for
|
||||
from artemis.services.llm import _record_llm_event, _record_llm_retry, get_llm
|
||||
from artemis.utils.logger import get_logger
|
||||
@@ -159,16 +145,8 @@ class VideoAnalyzer:
|
||||
getattr(agent_config, "enable_video_ledger", True),
|
||||
)
|
||||
self.static_chunk_size_seconds = float(
|
||||
getattr(self.video_config, "chunk_size_seconds", 60.0)
|
||||
getattr(self.video_config, "chunk_size_seconds", settings.ARTEMIS_VIDEO_CHUNK_SECONDS)
|
||||
)
|
||||
self.agentic_chunk_size_seconds = float(
|
||||
getattr(self.video_config, "agentic_chunk_size_seconds", 600.0)
|
||||
)
|
||||
self.agentic_call_timeout_seconds = float(
|
||||
getattr(self.video_config, "agentic_call_timeout_seconds", 300.0)
|
||||
)
|
||||
# Effective chunk ceiling; re-derived once the engine picks a
|
||||
# video-processing mode (agentic sub-agents take long clips).
|
||||
self.chunk_size_seconds = self.static_chunk_size_seconds
|
||||
self.video_processing = "static"
|
||||
self.min_chunk_seconds = float(getattr(self.video_config, "min_chunk_seconds", 4.0))
|
||||
@@ -178,9 +156,15 @@ class VideoAnalyzer:
|
||||
self.max_dense_action_frames = int(
|
||||
getattr(self.video_config, "max_dense_action_frames", 24)
|
||||
)
|
||||
self.native_max_retries = int(getattr(self.video_config, "native_max_retries", 1))
|
||||
self.max_retries = int(
|
||||
getattr(self.video_config, "max_retries", settings.ARTEMIS_VIDEO_MAX_RETRIES)
|
||||
)
|
||||
self.model_call_timeout_seconds = float(
|
||||
getattr(self.video_config, "model_call_timeout_seconds", 120.0)
|
||||
getattr(
|
||||
self.video_config,
|
||||
"model_call_timeout_seconds",
|
||||
settings.ARTEMIS_VIDEO_TIMEOUT_SECONDS,
|
||||
)
|
||||
)
|
||||
breaker = getattr(ctx, "_video_circuit_breaker", None)
|
||||
if breaker is None:
|
||||
@@ -210,7 +194,6 @@ class VideoAnalyzer:
|
||||
audio_prompt_path.read_text(encoding="utf-8") if audio_prompt_path.exists() else ""
|
||||
)
|
||||
|
||||
self._init_tools()
|
||||
self._init_engine()
|
||||
|
||||
def _get_universal_llm(self, *, use_fallback: bool = False):
|
||||
@@ -406,65 +389,13 @@ class VideoAnalyzer:
|
||||
analysis = match.group(2).strip() if match.group(2) else None
|
||||
return summary, analysis
|
||||
|
||||
def _init_tools(self) -> None:
|
||||
"""Initializes native and sub-agent tool declarations."""
|
||||
self.tools_declaration = build_native_tools_declaration()
|
||||
self.submit_answer_tool = build_submit_answer_declaration()
|
||||
|
||||
def _init_engine(self) -> None:
|
||||
"""Initializes model engine and decides whether to use native Gemini or Universal path."""
|
||||
ctx = self.ctx
|
||||
llm_config = getattr(ctx, "llm_config", None)
|
||||
llm_config = getattr(self.ctx, "llm_config", None)
|
||||
utils_cfg = getattr(llm_config, "utils", None) if llm_config else None
|
||||
llm_cfg = getattr(utils_cfg, "video_analyzer", None) if utils_cfg else None
|
||||
self.model_name = strip_provider_prefix(
|
||||
llm_cfg.model if (llm_cfg and hasattr(llm_cfg, "model")) else "gemini-3.8-flash"
|
||||
)
|
||||
|
||||
has_google_key = bool(
|
||||
settings.GOOGLE_API_KEY and settings.GOOGLE_API_KEY.get_secret_value()
|
||||
)
|
||||
|
||||
# If client is already set on ctx or explicit Gemini configuration
|
||||
self.client = getattr(ctx, "_genai_client", None)
|
||||
if self.client is not None:
|
||||
self.use_native_gemini = True
|
||||
elif has_google_key and is_gemini_model(self.model_name):
|
||||
try:
|
||||
self.client = genai.Client(api_key=settings.GOOGLE_API_KEY.get_secret_value())
|
||||
ctx._genai_client = self.client
|
||||
self.use_native_gemini = True
|
||||
asyncio.create_task(cleanup_abandoned_gemini_files(self.client))
|
||||
except Exception as e:
|
||||
logger.warning(
|
||||
f"Failed to initialize native Gemini client: {e}. Using universal engine."
|
||||
)
|
||||
self.use_native_gemini = False
|
||||
else:
|
||||
self.use_native_gemini = False
|
||||
self._resolve_video_processing()
|
||||
|
||||
def _resolve_video_processing(self) -> None:
|
||||
"""Picks static vs agentic video understanding for the sub-agents.
|
||||
|
||||
Agentic processing (Gemini Interactions API) needs the native client
|
||||
and a supporting model; the configured knob (``auto`` / ``agentic`` /
|
||||
``static``) decides the rest. The chunk ceiling follows the mode
|
||||
because agentic sub-agents are built to take long clips.
|
||||
"""
|
||||
configured = getattr(self.video_config, "processing", "auto")
|
||||
mode = "static"
|
||||
if self.use_native_gemini:
|
||||
mode = resolve_video_processing(configured, self.model_name)
|
||||
self.video_processing = mode
|
||||
self.chunk_size_seconds = (
|
||||
self.agentic_chunk_size_seconds if mode == "agentic" else self.static_chunk_size_seconds
|
||||
)
|
||||
if mode == "agentic":
|
||||
logger.info(
|
||||
f"Video sub-agents use agentic video understanding on {self.model_name}"
|
||||
f" (chunk ceiling {self.chunk_size_seconds:.0f}s)"
|
||||
)
|
||||
self.model_name = getattr(llm_cfg, "model", None) or settings.OPENAI_MODEL
|
||||
self.video_processing = "static"
|
||||
self.chunk_size_seconds = self.static_chunk_size_seconds
|
||||
|
||||
@property
|
||||
def blackboard_ledger(self) -> str:
|
||||
@@ -476,13 +407,6 @@ class VideoAnalyzer:
|
||||
lines.append(f"{e['start']}s - {e['end']}s: {e['summary']}")
|
||||
return " | ".join(lines)
|
||||
|
||||
async def upload_and_poll_file(self, compressed_path: Path) -> any:
|
||||
from artemis.agents.video_analyzer import gemini_files
|
||||
|
||||
return await gemini_files.upload_and_poll_file(
|
||||
self.client, compressed_path, self.cloud_files_to_cleanup
|
||||
)
|
||||
|
||||
def get_overlapping_warnings(self, start_time: float, end_time: float | None) -> list[dict]:
|
||||
warnings = []
|
||||
if not isinstance(start_time, (int, float)) or (
|
||||
@@ -739,17 +663,17 @@ class VideoAnalyzer:
|
||||
async def _exec_single_chunk(
|
||||
self, start_time: float, end_time: float | None, specific_query: str
|
||||
) -> str:
|
||||
from artemis.agents.video_analyzer import chunk_native
|
||||
from artemis.agents.video_analyzer import chunk_runner
|
||||
|
||||
return await chunk_native.exec_single_chunk(self, start_time, end_time, specific_query)
|
||||
return await chunk_runner.exec_single_chunk(self, start_time, end_time, specific_query)
|
||||
|
||||
@trace(type="tool", name="analyze_audio_only")
|
||||
async def exec_analyze_audio_only(
|
||||
self, start_time: float, end_time: float | None, specific_query: str
|
||||
) -> str:
|
||||
from artemis.agents.video_analyzer import audio_native
|
||||
from artemis.agents.video_analyzer import audio_runner
|
||||
|
||||
return await audio_native.exec_analyze_audio_only(
|
||||
return await audio_runner.exec_analyze_audio_only(
|
||||
self, start_time, end_time, specific_query
|
||||
)
|
||||
|
||||
@@ -837,15 +761,12 @@ class VideoAnalyzer:
|
||||
utils_cfg = getattr(llm_config, "utils", None) if llm_config else None
|
||||
llm_cfg = getattr(utils_cfg, "video_analyzer", None) if utils_cfg else None
|
||||
temperature = 0.2
|
||||
thinking_level = None
|
||||
if llm_cfg:
|
||||
self.model_name = strip_provider_prefix(llm_cfg.model)
|
||||
self.model_name = llm_cfg.model
|
||||
if getattr(llm_cfg, "temperature", None) is not None:
|
||||
temperature = llm_cfg.temperature
|
||||
if getattr(llm_cfg, "thinking_level", None) is not None:
|
||||
thinking_level = llm_cfg.thinking_level
|
||||
else:
|
||||
self.model_name = "gemini-3.8-flash"
|
||||
self.model_name = settings.OPENAI_MODEL
|
||||
|
||||
# Track files for cleanup
|
||||
self.local_files_to_cleanup = set()
|
||||
@@ -868,24 +789,11 @@ class VideoAnalyzer:
|
||||
" proceeding.\n"
|
||||
)
|
||||
|
||||
if not self.use_native_gemini:
|
||||
try:
|
||||
return await self._run_universal(
|
||||
time_description=time_description,
|
||||
purpose=purpose,
|
||||
system_prompt=system_prompt,
|
||||
)
|
||||
finally:
|
||||
await self._cleanup_universal_resources()
|
||||
|
||||
from artemis.agents.video_analyzer import native_coordinator
|
||||
|
||||
return await native_coordinator.run_native(
|
||||
self,
|
||||
ctx,
|
||||
time_description,
|
||||
purpose,
|
||||
system_prompt,
|
||||
temperature,
|
||||
thinking_level,
|
||||
)
|
||||
try:
|
||||
return await self._run_universal(
|
||||
time_description=time_description,
|
||||
purpose=purpose,
|
||||
system_prompt=system_prompt,
|
||||
)
|
||||
finally:
|
||||
await self._cleanup_universal_resources()
|
||||
Reference in new issue
Block a user