Add AI chat with skills management and model provider settings
- Model providers for OpenAI and Anthropic message formats (e.g. via Sub2API), with API keys encrypted at rest and never returned - Skills tree with markdown editing, priorities, drafts, publishing and rollback; only published skills enter the chat prompt - Chat page with tool calls over collected news and Binance USD-M klines, a persisted CONTEXT.md, reset, and stop-generation support - Chat may only create skill drafts and update CONTEXT.md; publishing, deleting, moving and priority changes stay with the user
This commit is contained in:
eezd committed
1 parent
f042f27013
commit
5db3ead883
34 files changed
+3196
-12
No files matched your search
@@ -0,0 +1,197 @@
|
||||
"""两种协议的模型适配：OpenAI Chat Completions 与 Anthropic Messages，统一成流式事件
|
||||
|
||||
消息使用与协议无关的格式（与 ChatMessage.content 一致，外加 role）：
|
||||
{"role": "user", "text"}
|
||||
{"role": "assistant", "text", "tool_calls": [{"id", "name", "arguments"}]}
|
||||
{"role": "tool", "tool_call_id", "name", "result"}
|
||||
"""
|
||||
|
||||
import json
|
||||
from collections.abc import AsyncIterator
|
||||
from dataclasses import dataclass, field
|
||||
from typing import Any
|
||||
|
||||
import anthropic
|
||||
import openai
|
||||
|
||||
from alphapulse.core.crypto import decrypt
|
||||
from alphapulse.models import AiProvider
|
||||
|
||||
TIMEOUT_SECONDS = 120
|
||||
# SDK 自带重试只针对连接错误、429 和 5xx
|
||||
MAX_RETRIES = 2
|
||||
|
||||
|
||||
class LlmError(Exception):
|
||||
"""模型调用失败；message 可直接展示给用户"""
|
||||
|
||||
|
||||
@dataclass
|
||||
class ToolSpec:
|
||||
name: str
|
||||
description: str
|
||||
# JSON Schema
|
||||
parameters: dict
|
||||
|
||||
|
||||
@dataclass
|
||||
class Completion:
|
||||
text: str = ""
|
||||
tool_calls: list[dict] = field(default_factory=list)
|
||||
usage: dict = field(default_factory=dict)
|
||||
|
||||
|
||||
@dataclass
|
||||
class TextDelta:
|
||||
text: str
|
||||
|
||||
|
||||
def api_root(base_url: str) -> str:
|
||||
"""用户可能填 https://host 或 https://host/v1，统一成不带 /v1 的根地址"""
|
||||
root = base_url.rstrip("/")
|
||||
return root.removesuffix("/v1")
|
||||
|
||||
|
||||
async def stream(provider: AiProvider, system: str, messages: list[dict], tools: list[ToolSpec]) -> AsyncIterator[TextDelta | Completion]:
|
||||
"""先逐段产出 TextDelta，最后产出一个 Completion"""
|
||||
try:
|
||||
if provider.protocol == "openai":
|
||||
async for event in _stream_openai(provider, system, messages, tools):
|
||||
yield event
|
||||
else:
|
||||
async for event in _stream_anthropic(provider, system, messages, tools):
|
||||
yield event
|
||||
except (openai.APIStatusError, anthropic.APIStatusError) as e:
|
||||
raise LlmError(f"模型接口返回 {e.status_code}：{e.message}") from e
|
||||
except (openai.APITimeoutError, anthropic.APITimeoutError) as e:
|
||||
raise LlmError(f"模型接口超时（{TIMEOUT_SECONDS}s）") from e
|
||||
except (openai.APIConnectionError, anthropic.APIConnectionError) as e:
|
||||
raise LlmError(f"无法连接模型接口：{e}") from e
|
||||
|
||||
|
||||
# ---------- OpenAI ----------
|
||||
|
||||
def _openai_messages(system: str, messages: list[dict]) -> list[dict]:
|
||||
out: list[dict] = [{"role": "system", "content": system}]
|
||||
for m in messages:
|
||||
if m["role"] == "user":
|
||||
out.append({"role": "user", "content": m["text"]})
|
||||
elif m["role"] == "assistant":
|
||||
if not m["text"] and not m.get("tool_calls"):
|
||||
# 停止得太早或模型没有输出：空的 assistant 消息会被接口拒绝
|
||||
continue
|
||||
msg: dict[str, Any] = {"role": "assistant", "content": m["text"] or None}
|
||||
if m.get("tool_calls"):
|
||||
msg["tool_calls"] = [
|
||||
{"id": c["id"], "type": "function", "function": {"name": c["name"], "arguments": json.dumps(c["arguments"], ensure_ascii=False)}}
|
||||
for c in m["tool_calls"]
|
||||
]
|
||||
out.append(msg)
|
||||
else:
|
||||
out.append({"role": "tool", "tool_call_id": m["tool_call_id"], "content": m["result"]})
|
||||
return out
|
||||
|
||||
|
||||
async def _stream_openai(provider: AiProvider, system: str, messages: list[dict], tools: list[ToolSpec]) -> AsyncIterator[TextDelta | Completion]:
|
||||
client = openai.AsyncOpenAI(
|
||||
api_key=decrypt(provider.api_key_encrypted), base_url=f"{api_root(provider.base_url)}/v1",
|
||||
timeout=TIMEOUT_SECONDS, max_retries=MAX_RETRIES,
|
||||
)
|
||||
kwargs: dict[str, Any] = {}
|
||||
if tools:
|
||||
kwargs["tools"] = [{"type": "function", "function": {"name": t.name, "description": t.description, "parameters": t.parameters}} for t in tools]
|
||||
if provider.temperature is not None:
|
||||
kwargs["temperature"] = provider.temperature
|
||||
result = Completion()
|
||||
# 按 index 拼接分片到达的工具调用
|
||||
calls: dict[int, dict] = {}
|
||||
async with client:
|
||||
chunks = await client.chat.completions.create(
|
||||
model=provider.model, messages=_openai_messages(system, messages), max_tokens=provider.max_tokens,
|
||||
stream=True, stream_options={"include_usage": True}, **kwargs,
|
||||
)
|
||||
async for chunk in chunks:
|
||||
if chunk.usage:
|
||||
result.usage = {"input_tokens": chunk.usage.prompt_tokens, "output_tokens": chunk.usage.completion_tokens}
|
||||
if not chunk.choices:
|
||||
continue
|
||||
delta = chunk.choices[0].delta
|
||||
if delta.content:
|
||||
result.text += delta.content
|
||||
yield TextDelta(delta.content)
|
||||
for tc in delta.tool_calls or []:
|
||||
call = calls.setdefault(tc.index, {"id": "", "name": "", "arguments": ""})
|
||||
if tc.id:
|
||||
call["id"] = tc.id
|
||||
if tc.function and tc.function.name:
|
||||
call["name"] += tc.function.name
|
||||
if tc.function and tc.function.arguments:
|
||||
call["arguments"] += tc.function.arguments
|
||||
result.tool_calls = [_parse_call(c["id"], c["name"], c["arguments"]) for _, c in sorted(calls.items())]
|
||||
yield result
|
||||
|
||||
|
||||
# ---------- Anthropic ----------
|
||||
|
||||
def _anthropic_messages(messages: list[dict]) -> list[dict]:
|
||||
out: list[dict] = []
|
||||
for m in messages:
|
||||
if m["role"] == "user":
|
||||
out.append({"role": "user", "content": m["text"]})
|
||||
elif m["role"] == "assistant":
|
||||
blocks: list[dict] = [{"type": "text", "text": m["text"]}] if m["text"] else []
|
||||
blocks += [{"type": "tool_use", "id": c["id"], "name": c["name"], "input": c["arguments"]} for c in m.get("tool_calls") or []]
|
||||
# 空的 assistant 消息会被接口拒绝
|
||||
if blocks:
|
||||
out.append({"role": "assistant", "content": blocks})
|
||||
else:
|
||||
block = {"type": "tool_result", "tool_use_id": m["tool_call_id"], "content": m["result"]}
|
||||
# 同一轮的多个工具结果必须放进同一条 user 消息
|
||||
if out and out[-1]["role"] == "user" and isinstance(out[-1]["content"], list):
|
||||
out[-1]["content"].append(block)
|
||||
else:
|
||||
out.append({"role": "user", "content": [block]})
|
||||
return out
|
||||
|
||||
|
||||
async def _stream_anthropic(provider: AiProvider, system: str, messages: list[dict], tools: list[ToolSpec]) -> AsyncIterator[TextDelta | Completion]:
|
||||
client = anthropic.AsyncAnthropic(
|
||||
api_key=decrypt(provider.api_key_encrypted), base_url=api_root(provider.base_url),
|
||||
timeout=TIMEOUT_SECONDS, max_retries=MAX_RETRIES,
|
||||
)
|
||||
kwargs: dict[str, Any] = {}
|
||||
if tools:
|
||||
kwargs["tools"] = [{"name": t.name, "description": t.description, "input_schema": t.parameters} for t in tools]
|
||||
if provider.temperature is not None:
|
||||
# 当前 SDK 不再提供 temperature 参数，按原始字段透传
|
||||
kwargs["extra_body"] = {"temperature": provider.temperature}
|
||||
result = Completion()
|
||||
blocks: dict[int, dict] = {}
|
||||
async with client:
|
||||
events = await client.messages.create(
|
||||
model=provider.model, system=system, messages=_anthropic_messages(messages), max_tokens=provider.max_tokens,
|
||||
stream=True, **kwargs,
|
||||
)
|
||||
async for event in events:
|
||||
if event.type == "message_start":
|
||||
result.usage["input_tokens"] = event.message.usage.input_tokens
|
||||
elif event.type == "content_block_start" and event.content_block.type == "tool_use":
|
||||
blocks[event.index] = {"id": event.content_block.id, "name": event.content_block.name, "arguments": ""}
|
||||
elif event.type == "content_block_delta":
|
||||
if event.delta.type == "text_delta":
|
||||
result.text += event.delta.text
|
||||
yield TextDelta(event.delta.text)
|
||||
elif event.delta.type == "input_json_delta":
|
||||
blocks[event.index]["arguments"] += event.delta.partial_json
|
||||
elif event.type == "message_delta":
|
||||
result.usage["output_tokens"] = event.usage.output_tokens
|
||||
result.tool_calls = [_parse_call(b["id"], b["name"], b["arguments"]) for _, b in sorted(blocks.items())]
|
||||
yield result
|
||||
|
||||
|
||||
def _parse_call(call_id: str, name: str, arguments: str) -> dict:
|
||||
try:
|
||||
parsed = json.loads(arguments) if arguments else {}
|
||||
except json.JSONDecodeError as e:
|
||||
raise LlmError(f"模型返回的工具参数不是合法 JSON：{name} {arguments[:200]}") from e
|
||||
return {"id": call_id, "name": name, "arguments": parsed}
|
||||
Reference in new issue
Block a user