mirror of
https://gitcode.com/JianFeeeee/ModelRouter.git
synced 2026-11-07 09:47:23 +00:00
93eeba3762
Phase 2 收尾（入向）。ChatRequest 从"客户端发什么就是什么"改为
UnifiedRequest（最大集合）的别名，内核真正用上了那个超集类型。
改掉的具体缺陷：
1. 私有字段泄漏归零（v1 的核心问题）
v1 的 disable_thinking 是内核私有开关，却被序列化进交给适配器的 body，
迫使 10 个适配器各自写 req.disable_thinking = nil 来清理。
现在它是中立的 thinking 字段（{enabled, budget_tokens}），客户端与
适配器都认这个词：
- gateway/chat.go：thinkingFromClient() 把 disable_thinking 翻译成
thinking.enabled=false；冲突时中立字段优先（它能带预算，布尔不能）
- deepseek.lua：thinking.enabled=false → extra_body.thinking.type=disabled
- anthropic.lua：thinking.enabled=true + budget → 原生 {type,budget_tokens}
（enabled=false 时不传该字段——Anthropic 没有 type=disabled 这个写法）
- 其余 7 个适配器：把"清理内核私有字段"改成"清理中立字段"
2. Tools/ToolChoice 改 json.RawMessage
原为 []interface{}（OpenAI 原始形状 = 内核在假设某个协议）。
内核从不解释工具 schema——它原样交给适配器，所以原始形状不损失任何
内核侧统一性。ToolDef 结构化类型保留在包内供新适配器选用。
3. 删除 MarshalJSON 后门
v1 手工 splice ExtraBody 进 OpenAI 形状的 body；现在 ExtraBody 是
UnifiedRequest 上一个正正经经声明的字段。
渐进迁移：ChatRequest 保留为别名（不是新类型），49 处引用零改动编译，
语义变更不被机械改名淹没。
测试：新增 thinking_test.go（翻译链 4 例 + 泄漏回归 + tools 透传）；
vm_test.go 两处迁移到中立语义。golden 字节级未漂移。
293 lines
12 KiB
Lua
293 lines
12 KiB
Lua
-- trae 源适配器（trae-local-api 的 OpenAI 兼容代理）
|
||
-- trae-local-api 运行在 http://127.0.0.1:19900，接 Trae CN 账号的云资源。
|
||
-- 协议：OpenAI /v1/chat/completions
|
||
-- 特性：
|
||
-- - 排队：上游 busy 时排队，可能长时间无响应
|
||
-- - 非流式请求偶发返回 SSE 数据（upstream bug）
|
||
-- - 支持 thinking，由上游模型自动决定是否开启
|
||
|
||
local adapter = {}
|
||
adapter.name = "trae"
|
||
adapter.version = "1.0.0"
|
||
adapter.endpoint = "/v1/chat/completions"
|
||
adapter.headers = {}
|
||
|
||
function adapter.transform_request(raw_body)
|
||
local ok, req = pcall(json.decode, raw_body)
|
||
if not ok then return raw_body end
|
||
-- The gateway's private control fields are GONE from the body (v2): the
|
||
-- neutral `thinking` field carries that intent and each adapter maps it in
|
||
-- its own protocol terms. These upstreams have no thinking switch at all,
|
||
-- so the field is simply dropped here — which is what the nil-ing below
|
||
-- used to do, except the field being dropped is no longer the kernel's
|
||
-- private one.
|
||
req.thinking = nil
|
||
req.extra_body = nil
|
||
if req.messages then
|
||
for _, msg in ipairs(req.messages) do
|
||
msg.reasoning_content = nil
|
||
end
|
||
end
|
||
return json.encode(req)
|
||
end
|
||
|
||
-- parse_text_tool_calls extracts tool calls that an upstream emitted as PLAIN
|
||
-- TEXT instead of using the OpenAI tool_calls field.
|
||
--
|
||
-- trae-local-api's OpenAI endpoint (/v1/chat/completions) does not read the
|
||
-- request's `tools` array at all, so the relayed model is never told the tool
|
||
-- schema; it falls back to printing
|
||
-- <tool_call>
|
||
-- {"name": "get_weather", "arguments": {"location": "北京"}}
|
||
-- </tool_call>
|
||
-- into message.content, leaves message.tool_calls null, and reports
|
||
-- finish_reason="stop". A client following the OpenAI contract therefore never
|
||
-- sees a tool call: the agent loop terminates unexpectedly mid-conversation
|
||
-- (and a hand-written replay produces an empty function name next turn).
|
||
--
|
||
-- Both tag spellings are accepted: the same codebase's Anthropic endpoint
|
||
-- instructs models to emit <toolcall>, and models mix the two. Key names vary
|
||
-- too (arguments / params / input), so all are tried.
|
||
--
|
||
-- Returns (tool_calls_array_or_nil, content_with_blocks_removed).
|
||
local function parse_text_tool_calls(content)
|
||
if type(content) ~= "string" or content == "" then return nil, content end
|
||
|
||
local tcs = {}
|
||
local idx = 0
|
||
|
||
local function collect(pattern)
|
||
for payload in content:gmatch(pattern) do
|
||
local ok, obj = pcall(json.decode, payload)
|
||
if ok and type(obj) == "table" then
|
||
-- some models wrap it as {"function":{"name":..,"arguments":..}}
|
||
local fn = obj["function"]
|
||
local name = obj.name or (type(fn) == "table" and fn.name) or nil
|
||
if name then
|
||
local args = obj.arguments or obj.params or obj.input
|
||
if args == nil and type(fn) == "table" then
|
||
args = fn.arguments or fn.params
|
||
end
|
||
if type(args) == "string" then
|
||
local aok, decoded = pcall(json.decode, args)
|
||
args = aok and decoded or {}
|
||
elseif type(args) ~= "table" then
|
||
args = {}
|
||
end
|
||
idx = idx + 1
|
||
table.insert(tcs, {
|
||
id = obj.id or ("call_text_" .. idx),
|
||
type = "function",
|
||
name = name,
|
||
arguments = args
|
||
})
|
||
end
|
||
end
|
||
end
|
||
end
|
||
|
||
-- `<tag ...>` allows attributes; %s* handles the "</tool_call >" spacing
|
||
-- that trae-local-api's own prompt example uses.
|
||
collect("<tool_call[^>]*>%s*(.-)%s*</tool_call%s*>")
|
||
collect("<toolcall[^>]*>%s*(.-)%s*</toolcall%s*>")
|
||
|
||
-- Legacy / mimicked form: models that saw past assistant turns folded as
|
||
-- [Called tool: name({...})]
|
||
-- tend to emit the same shape instead of a real block. Recover it so
|
||
-- the agent loop keeps working; otherwise the client receives plain text
|
||
-- where a structured tool call should be. Capture the name and the JSON
|
||
-- object literal independently (the JSON is the only {...} run here).
|
||
if #tcs == 0 and content:find("%[Called tool") then
|
||
for name, argsraw in content:gmatch("%[Called tool%s*:%s*([%w_%-%.]+)%s*%((%b{})%s*%)%s*%]") do
|
||
local aok, decoded = pcall(json.decode, argsraw)
|
||
idx = idx + 1
|
||
table.insert(tcs, {
|
||
id = "call_legacy_" .. idx,
|
||
type = "function",
|
||
name = name,
|
||
arguments = aok and decoded or {}
|
||
})
|
||
end
|
||
end
|
||
|
||
if #tcs == 0 then return nil, content end
|
||
|
||
-- drop the blocks from user-visible content; keep any surrounding prose
|
||
local stripped = content:gsub("<tool_call[^>]*>%s*.-%s*</tool_call%s*>", "")
|
||
stripped = stripped:gsub("<toolcall[^>]*>%s*.-%s*</toolcall%s*>", "")
|
||
stripped = stripped:gsub("%[Called tool%s*:%s*[%w_%-%.]+%s*%(%b{}%s*%)%s*%]", "")
|
||
stripped = stripped:gsub("^%s+", ""):gsub("%s+$", "")
|
||
return tcs, stripped
|
||
end
|
||
|
||
function adapter.transform_response(raw_body)
|
||
-- trae-local-api 偶发在非流式请求中返回 SSE 格式数据，
|
||
-- 表现为多个 data: {...} 行或混合了 reasoning_chunk 等。
|
||
-- 剥掉 data: 前缀，取最后一个完整的 JSON 块（通常是
|
||
-- 最终的 stop chunk 或 usage chunk）。
|
||
local use_json = raw_body
|
||
if raw_body:match("^data:") or raw_body:match("\ndata:") then
|
||
-- 取最后一个 data: 行的 JSON
|
||
local last = nil
|
||
for line in raw_body:gmatch("data: ([^\n\r]+)") do
|
||
local trimmed = line:match("^%s*(.-)%s*$")
|
||
if trimmed and trimmed ~= "[DONE]" then
|
||
local ok2, obj = pcall(json.decode, trimmed)
|
||
if ok2 and type(obj) == "table" then
|
||
last = obj
|
||
end
|
||
end
|
||
end
|
||
if last then
|
||
use_json = json.encode(last)
|
||
end
|
||
-- 如果解析失败，回退到原始 body
|
||
end
|
||
|
||
local ok, resp = pcall(json.decode, use_json)
|
||
if not ok or resp == nil then return raw_body end
|
||
|
||
local unified = {
|
||
content = "",
|
||
finish_reason = "",
|
||
token_usage = { prompt = 0, completion = 0, total = 0 }
|
||
}
|
||
|
||
if type(resp.usage) == "table" then
|
||
unified.token_usage.prompt = resp.usage.prompt_tokens or 0
|
||
unified.token_usage.completion = resp.usage.completion_tokens or 0
|
||
unified.token_usage.total = resp.usage.total_tokens or 0
|
||
end
|
||
|
||
if type(resp.choices) == "table" and #resp.choices > 0 then
|
||
local ch = resp.choices[1]
|
||
if type(ch.message) == "table" then
|
||
unified.content = ch.message.content or ""
|
||
if ch.message.reasoning_content then
|
||
unified.reasoning_content = ch.message.reasoning_content
|
||
end
|
||
if type(ch.message.tool_calls) == "table" and #ch.message.tool_calls > 0 then
|
||
local tcs = {}
|
||
for _, tc in ipairs(ch.message.tool_calls) do
|
||
local fn = tc["function"] or {}
|
||
local args = fn.arguments
|
||
if type(args) == "string" then
|
||
local args_ok, decoded = pcall(json.decode, args)
|
||
args = args_ok and decoded or {}
|
||
elseif type(args) ~= "table" then
|
||
args = {}
|
||
end
|
||
table.insert(tcs, {
|
||
id = tc.id,
|
||
type = tc.type or "function",
|
||
name = fn.name,
|
||
arguments = args
|
||
})
|
||
end
|
||
unified.tool_calls = tcs
|
||
end
|
||
end
|
||
unified.finish_reason = ch.finish_reason or ""
|
||
|
||
-- Fallback: trae-local-api's OpenAI endpoint drops the request's
|
||
-- `tools` array entirely, so the relayed model is never told the tool
|
||
-- schema and instead PRINTS a <tool_call>{...}</tool_call> block into
|
||
-- content, leaving message.tool_calls null and finish_reason="stop".
|
||
-- A client following the OpenAI contract then sees a normal completion
|
||
-- and its agent loop terminates mid-conversation. Recover the
|
||
-- structured call so the loop can continue.
|
||
if unified.tool_calls == nil then
|
||
local recovered, cleaned = parse_text_tool_calls(unified.content)
|
||
if recovered then
|
||
unified.tool_calls = recovered
|
||
unified.content = cleaned
|
||
unified.finish_reason = "tool_calls"
|
||
end
|
||
end
|
||
end
|
||
|
||
return json.encode(unified)
|
||
end
|
||
|
||
function adapter.transform_stream_chunk(raw_chunk)
|
||
local ok, chunk = pcall(json.decode, raw_chunk)
|
||
if not ok then return "" end
|
||
|
||
local uses = nil
|
||
if type(chunk.usage) == "table" then
|
||
uses = {
|
||
prompt = chunk.usage.prompt_tokens or chunk.usage.prompt or 0,
|
||
completion = chunk.usage.completion_tokens or chunk.usage.completion or 0,
|
||
total = chunk.usage.total_tokens or chunk.usage.total or 0,
|
||
}
|
||
if type(chunk.usage.prompt_tokens_details) == "table" and chunk.usage.prompt_tokens_details.cached_tokens ~= nil then
|
||
uses.prompt_tokens_details = { cached_tokens = chunk.usage.prompt_tokens_details.cached_tokens }
|
||
elseif (chunk.usage.prompt_cache_hit_tokens or 0) > 0 then
|
||
uses.prompt_cache_hit_tokens = chunk.usage.prompt_cache_hit_tokens
|
||
uses.prompt_cache_miss_tokens = chunk.usage.prompt_cache_miss_tokens or 0
|
||
uses.prompt_tokens_details = { cached_tokens = chunk.usage.prompt_cache_hit_tokens }
|
||
end
|
||
end
|
||
|
||
if not chunk.choices or #chunk.choices == 0 then
|
||
if uses ~= nil then
|
||
return json.encode({ usage = uses, done = false })
|
||
end
|
||
return ""
|
||
end
|
||
local delta = chunk.choices[1].delta or {}
|
||
local fr = chunk.choices[1].finish_reason
|
||
|
||
local finish = (type(fr) == "string" and fr ~= "") and fr or nil
|
||
|
||
local unified = {
|
||
content = delta.content or "",
|
||
done = (finish ~= nil)
|
||
}
|
||
if finish then
|
||
unified.finish_reason = finish
|
||
end
|
||
if uses ~= nil then
|
||
unified.usage = uses
|
||
end
|
||
if delta.reasoning_content then
|
||
unified.reasoning_content = delta.reasoning_content
|
||
end
|
||
if delta.tool_calls then
|
||
unified.tool_calls = delta.tool_calls
|
||
end
|
||
return json.encode(unified)
|
||
end
|
||
|
||
function adapter.transform_error(status, body)
|
||
local ok, resp = pcall(json.decode, body)
|
||
if not ok or type(resp) ~= "table" then
|
||
-- Non-JSON body: extract title from HTML or first line
|
||
local title = string.match(body or "", "<title>(.-)</title>")
|
||
if title and title ~= "" then return title end
|
||
local line = string.match(body or "", "^%s*([^\r\n]+)")
|
||
if line and line ~= "" and not string.match(line, "^<") then
|
||
return string.sub(line, 1, 200)
|
||
end
|
||
return nil
|
||
end
|
||
local e = resp.error
|
||
if type(e) == "table" and type(e.message) == "string" then
|
||
-- trae quota / rate limit annotations
|
||
if e.code then
|
||
return tostring(e.code) .. ": " .. e.message
|
||
end
|
||
return e.message
|
||
end
|
||
if type(e) == "string" then return e end
|
||
if type(resp.message) == "string" and resp.message ~= "" then
|
||
if resp.code ~= nil then
|
||
return tostring(resp.code) .. ": " .. resp.message
|
||
end
|
||
return resp.message
|
||
end
|
||
return nil
|
||
end
|
||
|
||
return adapter
|