Kris
1a7690aaa0
新增了完整的流式响应处理模块，包含代码围栏解析、分块策略、指令处理、流式草稿循环等核心能力，支持按段落/句子等规则切割响应块，支持进度更新与指令回调，完善流式交互的全流程支持
112 lines
3.9 KiB
Python
112 lines
3.9 KiB
Python
from __future__ import annotations
|
|
|
|
from dataclasses import dataclass
|
|
from enum import StrEnum
|
|
|
|
from yuxi.channel.config.defaults import (
|
|
BLOCK_STREAMING_BREAK,
|
|
BLOCK_STREAMING_CHUNK_BREAK_PREFERENCE,
|
|
BLOCK_STREAMING_CHUNK_MAX_CHARS,
|
|
BLOCK_STREAMING_CHUNK_MIN_CHARS,
|
|
BLOCK_STREAMING_COALESCE_IDLE_MS,
|
|
BLOCK_STREAMING_COALESCE_JOINER,
|
|
BLOCK_STREAMING_COALESCE_MAX_CHARS,
|
|
BLOCK_STREAMING_COALESCE_MIN_CHARS,
|
|
STREAMING_PREVIEW_MIN_INITIAL_CHARS,
|
|
STREAMING_PREVIEW_THROTTLE_MS,
|
|
)
|
|
|
|
|
|
class BlockStreamingBreak(StrEnum):
|
|
TEXT_END = "text_end"
|
|
|
|
|
|
class DraftStreamState(StrEnum):
|
|
IDLE = "idle"
|
|
STREAMING = "streaming"
|
|
FINALIZING = "finalizing"
|
|
FINISHED = "finished"
|
|
ABORTED = "aborted"
|
|
|
|
|
|
@dataclass
|
|
class StreamingConfig:
|
|
preview_stream_throttle_ms: int = STREAMING_PREVIEW_THROTTLE_MS
|
|
preview_min_initial_chars: int = STREAMING_PREVIEW_MIN_INITIAL_CHARS
|
|
|
|
block_streaming_enabled: bool = False
|
|
block_streaming_break: str = BLOCK_STREAMING_BREAK
|
|
block_streaming_chunk_min_chars: int = BLOCK_STREAMING_CHUNK_MIN_CHARS
|
|
block_streaming_chunk_max_chars: int = BLOCK_STREAMING_CHUNK_MAX_CHARS
|
|
block_streaming_chunk_break_preference: str = BLOCK_STREAMING_CHUNK_BREAK_PREFERENCE
|
|
|
|
block_streaming_flush_on_paragraph: bool = False
|
|
|
|
block_streaming_coalesce_min_chars: int | None = BLOCK_STREAMING_COALESCE_MIN_CHARS
|
|
block_streaming_coalesce_max_chars: int | None = BLOCK_STREAMING_COALESCE_MAX_CHARS
|
|
block_streaming_coalesce_idle_ms: int = BLOCK_STREAMING_COALESCE_IDLE_MS
|
|
block_streaming_coalesce_joiner: str = BLOCK_STREAMING_COALESCE_JOINER
|
|
|
|
@classmethod
|
|
def defaults(cls) -> dict:
|
|
return {
|
|
"preview_stream_throttle_ms": STREAMING_PREVIEW_THROTTLE_MS,
|
|
"preview_min_initial_chars": STREAMING_PREVIEW_MIN_INITIAL_CHARS,
|
|
"block_streaming_enabled": False,
|
|
"block_streaming_break": BLOCK_STREAMING_BREAK,
|
|
"block_streaming_chunk_min_chars": BLOCK_STREAMING_CHUNK_MIN_CHARS,
|
|
"block_streaming_chunk_max_chars": BLOCK_STREAMING_CHUNK_MAX_CHARS,
|
|
"block_streaming_chunk_break_preference": BLOCK_STREAMING_CHUNK_BREAK_PREFERENCE,
|
|
"block_streaming_flush_on_paragraph": False,
|
|
"block_streaming_coalesce_min_chars": BLOCK_STREAMING_COALESCE_MIN_CHARS,
|
|
"block_streaming_coalesce_max_chars": BLOCK_STREAMING_COALESCE_MAX_CHARS,
|
|
"block_streaming_coalesce_idle_ms": BLOCK_STREAMING_COALESCE_IDLE_MS,
|
|
"block_streaming_coalesce_joiner": BLOCK_STREAMING_COALESCE_JOINER,
|
|
}
|
|
|
|
@classmethod
|
|
def empty(cls) -> StreamingConfig:
|
|
return cls()
|
|
|
|
|
|
@dataclass
|
|
class DraftStreamSession:
|
|
message_id: str | None = None
|
|
throttle_ms: int = STREAMING_PREVIEW_THROTTLE_MS
|
|
significant_delta_chars: int = STREAMING_PREVIEW_MIN_INITIAL_CHARS
|
|
state: DraftStreamState = DraftStreamState.IDLE
|
|
accumulated_text: str = ""
|
|
last_sent_text: str = ""
|
|
last_update_ts: float = 0.0
|
|
tool_status: str | None = None
|
|
|
|
stopped: bool = False
|
|
final: bool = False
|
|
|
|
@property
|
|
def is_active(self) -> bool:
|
|
return self.state in (DraftStreamState.STREAMING, DraftStreamState.FINALIZING)
|
|
|
|
@property
|
|
def pending_chars(self) -> str:
|
|
if len(self.accumulated_text) > len(self.last_sent_text):
|
|
return self.accumulated_text[len(self.last_sent_text) :]
|
|
return ""
|
|
|
|
|
|
@dataclass
|
|
class BlockReplyChunking:
|
|
min_chars: int = BLOCK_STREAMING_CHUNK_MIN_CHARS
|
|
max_chars: int = BLOCK_STREAMING_CHUNK_MAX_CHARS
|
|
break_preference: str = BLOCK_STREAMING_CHUNK_BREAK_PREFERENCE
|
|
flush_on_paragraph: bool = False
|
|
|
|
|
|
@dataclass
|
|
class BlockReplyCoalescing:
|
|
min_chars: int = BLOCK_STREAMING_CHUNK_MIN_CHARS
|
|
max_chars: int = BLOCK_STREAMING_CHUNK_MAX_CHARS
|
|
idle_ms: int = BLOCK_STREAMING_COALESCE_IDLE_MS
|
|
joiner: str = BLOCK_STREAMING_COALESCE_JOINER
|
|
flush_on_enqueue: bool = False
|