feat(llm): 思考开关与思考强度拆成两个独立可配置参数
- thinking_type(''/disabled/enabled) 与 reasoning_effort(''/low/medium/high/xhigh/max)
彻底解耦，可任意组合：支持「不发 thinking 字段、只靠强度激活思考」的接口
- agent.Config 拆为 ThinkingType + ReasoningEffort 两字段，NewProvider 直接透传
- db/schema.sql 加 thinking_type 列 + 一次性 DO 块迁移(仅列不存在时回填旧语义)
- 前端 LLM 配置页改为两个独立下拉，强度新增 xhigh 档
Claude-Session: https://claude.ai/code/session_01Bs2rrsVHUBTxRusm9espwB
This commit is contained in:
Autumn-27 committed
1 parent
834e94535a
commit
863e44ec24
9 files changed
+121
-83
No files matched your search
+15
-13
@@ -25,9 +25,11 @@ type LLMProfile struct {
|
||||
// ContextWindowK is the model's context window in K tokens, used to size
|
||||
// compaction thresholds. 0 = use a 200K default; capped at 1000 (1M).
|
||||
ContextWindowK int `json:"context_window_k"`
|
||||
// ReasoningEffort is the thinking-mode selector. "" = 默认(不发送思考参数);
|
||||
// "off" = 显式关闭(thinking.type=disabled); "low"/"medium"/"high"/"max" =
|
||||
// 开启思考并设强度. Mapped to the provider request in agent.Config.NewProvider.
|
||||
// ThinkingType 独立控制思考「开关」(thinking.type):"" = 不发送(默认);
|
||||
// "disabled" = 显式关闭; "enabled" = 开启. 与 ReasoningEffort 解耦.
|
||||
ThinkingType string `json:"thinking_type"`
|
||||
// ReasoningEffort 独立控制思考「强度」:"" = 不发送(默认);
|
||||
// "low"/"medium"/"high"/"xhigh"/"max" = 对应强度. 见 agent.Config.NewProvider.
|
||||
ReasoningEffort string `json:"reasoning_effort"`
|
||||
IsDefault bool `json:"is_default"`
|
||||
// Priority orders the failover chain: higher goes first. The ACTIVE profile
|
||||
@@ -41,14 +43,14 @@ type LLMProfile struct {
|
||||
|
||||
// profileCols is the read column list (hint variant, no api key) shared by the
|
||||
// list query; profileColsKey is the same with api_key for the single-row loads.
|
||||
const profileCols = `id,name,format,COALESCE(base_url,''),COALESCE(proxy,''),model,COALESCE(api_key_hint,''),rate_per_second,rate_per_minute,context_window_k,COALESCE(reasoning_effort,''),is_default,priority,pool_exclude`
|
||||
const profileColsKey = `id,name,format,COALESCE(base_url,''),COALESCE(proxy,''),model,COALESCE(api_key,''),rate_per_second,rate_per_minute,context_window_k,COALESCE(reasoning_effort,''),is_default,priority,pool_exclude`
|
||||
const profileCols = `id,name,format,COALESCE(base_url,''),COALESCE(proxy,''),model,COALESCE(api_key_hint,''),rate_per_second,rate_per_minute,context_window_k,COALESCE(reasoning_effort,''),is_default,priority,pool_exclude,COALESCE(thinking_type,'')`
|
||||
const profileColsKey = `id,name,format,COALESCE(base_url,''),COALESCE(proxy,''),model,COALESCE(api_key,''),rate_per_second,rate_per_minute,context_window_k,COALESCE(reasoning_effort,''),is_default,priority,pool_exclude,COALESCE(thinking_type,'')`
|
||||
|
||||
// scanProfile reads one row in the profileCols / profileColsKey column order. The
|
||||
// 7th column lands in APIKeyHint or APIKey depending on which list the caller used.
|
||||
func scanProfile(sc interface{ Scan(...any) error }, into *string, p *LLMProfile) error {
|
||||
return sc.Scan(&p.ID, &p.Name, &p.Format, &p.BaseURL, &p.Proxy, &p.Model, into,
|
||||
&p.RatePerSecond, &p.RatePerMinute, &p.ContextWindowK, &p.ReasoningEffort, &p.IsDefault, &p.Priority, &p.PoolExclude)
|
||||
&p.RatePerSecond, &p.RatePerMinute, &p.ContextWindowK, &p.ReasoningEffort, &p.IsDefault, &p.Priority, &p.PoolExclude, &p.ThinkingType)
|
||||
}
|
||||
|
||||
func (d *DB) ListProfiles() ([]*LLMProfile, error) {
|
||||
@@ -121,18 +123,18 @@ func (d *DB) SaveProfile(p *LLMProfile) (int64, error) {
|
||||
}
|
||||
if p.ID == 0 {
|
||||
var id int64
|
||||
err := d.QueryRow(`INSERT INTO llm_profiles(name,format,base_url,proxy,model,api_key,api_key_hint,rate_per_second,rate_per_minute,context_window_k,reasoning_effort,priority,pool_exclude)
|
||||
VALUES ($1,$2,NULLIF($3,''),NULLIF($4,''),$5,NULLIF($6,''),NULLIF($7,''),$8,$9,$10,$11,$12,$13) RETURNING id`,
|
||||
p.Name, p.Format, p.BaseURL, p.Proxy, p.Model, p.APIKey, hint, p.RatePerSecond, p.RatePerMinute, p.ContextWindowK, p.ReasoningEffort, p.Priority, p.PoolExclude).Scan(&id)
|
||||
err := d.QueryRow(`INSERT INTO llm_profiles(name,format,base_url,proxy,model,api_key,api_key_hint,rate_per_second,rate_per_minute,context_window_k,reasoning_effort,priority,pool_exclude,thinking_type)
|
||||
VALUES ($1,$2,NULLIF($3,''),NULLIF($4,''),$5,NULLIF($6,''),NULLIF($7,''),$8,$9,$10,$11,$12,$13,$14) RETURNING id`,
|
||||
p.Name, p.Format, p.BaseURL, p.Proxy, p.Model, p.APIKey, hint, p.RatePerSecond, p.RatePerMinute, p.ContextWindowK, p.ReasoningEffort, p.Priority, p.PoolExclude, p.ThinkingType).Scan(&id)
|
||||
return id, err
|
||||
}
|
||||
if p.APIKey == "" {
|
||||
_, err := d.Exec(`UPDATE llm_profiles SET name=$1,format=$2,base_url=NULLIF($3,''),proxy=NULLIF($4,''),model=$5,rate_per_second=$6,rate_per_minute=$7,context_window_k=$8,reasoning_effort=$9,priority=$10,pool_exclude=$11 WHERE id=$12`,
|
||||
p.Name, p.Format, p.BaseURL, p.Proxy, p.Model, p.RatePerSecond, p.RatePerMinute, p.ContextWindowK, p.ReasoningEffort, p.Priority, p.PoolExclude, p.ID)
|
||||
_, err := d.Exec(`UPDATE llm_profiles SET name=$1,format=$2,base_url=NULLIF($3,''),proxy=NULLIF($4,''),model=$5,rate_per_second=$6,rate_per_minute=$7,context_window_k=$8,reasoning_effort=$9,priority=$10,pool_exclude=$11,thinking_type=$12 WHERE id=$13`,
|
||||
p.Name, p.Format, p.BaseURL, p.Proxy, p.Model, p.RatePerSecond, p.RatePerMinute, p.ContextWindowK, p.ReasoningEffort, p.Priority, p.PoolExclude, p.ThinkingType, p.ID)
|
||||
return p.ID, err
|
||||
}
|
||||
_, err := d.Exec(`UPDATE llm_profiles SET name=$1,format=$2,base_url=NULLIF($3,''),proxy=NULLIF($4,''),model=$5,api_key=$6,api_key_hint=$7,rate_per_second=$8,rate_per_minute=$9,context_window_k=$10,reasoning_effort=$11,priority=$12,pool_exclude=$13 WHERE id=$14`,
|
||||
p.Name, p.Format, p.BaseURL, p.Proxy, p.Model, p.APIKey, hint, p.RatePerSecond, p.RatePerMinute, p.ContextWindowK, p.ReasoningEffort, p.Priority, p.PoolExclude, p.ID)
|
||||
_, err := d.Exec(`UPDATE llm_profiles SET name=$1,format=$2,base_url=NULLIF($3,''),proxy=NULLIF($4,''),model=$5,api_key=$6,api_key_hint=$7,rate_per_second=$8,rate_per_minute=$9,context_window_k=$10,reasoning_effort=$11,priority=$12,pool_exclude=$13,thinking_type=$14 WHERE id=$15`,
|
||||
p.Name, p.Format, p.BaseURL, p.Proxy, p.Model, p.APIKey, hint, p.RatePerSecond, p.RatePerMinute, p.ContextWindowK, p.ReasoningEffort, p.Priority, p.PoolExclude, p.ThinkingType, p.ID)
|
||||
return p.ID, err
|
||||
}
|
||||
|
||||
|
||||
@@ -235,7 +235,10 @@ CREATE TABLE IF NOT EXISTS llm_profiles (
|
||||
rate_per_second DOUBLE PRECISION NOT NULL DEFAULT 0,
|
||||
rate_per_minute DOUBLE PRECISION NOT NULL DEFAULT 0,
|
||||
context_window_k INTEGER NOT NULL DEFAULT 0,
|
||||
-- 思考参数拆成两个独立字段：thinking_type=思考开关(''/disabled/enabled)，
|
||||
-- reasoning_effort=思考强度(''/low/medium/high/xhigh/max)，互不牵连。
|
||||
reasoning_effort TEXT NOT NULL DEFAULT '',
|
||||
thinking_type TEXT NOT NULL DEFAULT '',
|
||||
is_default BOOLEAN NOT NULL DEFAULT false,
|
||||
-- 轮询(故障转移)参数，见 docs/LLM轮询设计.md：
|
||||
-- priority 顺位，越大越先被选中；激活配置(is_default)永远排链首，与本值无关。
|
||||
@@ -253,6 +256,26 @@ CREATE TRIGGER trg_llm_upd BEFORE UPDATE ON llm_profiles
|
||||
ALTER TABLE llm_profiles ADD COLUMN IF NOT EXISTS priority INTEGER NOT NULL DEFAULT 0;
|
||||
ALTER TABLE llm_profiles ADD COLUMN IF NOT EXISTS pool_exclude BOOLEAN NOT NULL DEFAULT false;
|
||||
|
||||
-- 思考开关字段 thinking_type，从旧的单一 reasoning_effort 语义一次性拆分而来。
|
||||
-- schema.sql 每次启动都执行，故迁移必须只跑一次：仅当该列尚不存在时才回填，
|
||||
-- 否则每次启动都会把用户后来手动设的组合覆盖回去。旧 reasoning_effort 语义：
|
||||
-- 'off' → 显式关闭 → thinking_type='disabled'，强度清空
|
||||
-- 'low/medium/high/max' → 开启+强度 → thinking_type='enabled'，强度保留
|
||||
-- '' → 不发送 → 两者皆空(默认)
|
||||
DO $$
|
||||
BEGIN
|
||||
IF NOT EXISTS (
|
||||
SELECT 1 FROM information_schema.columns
|
||||
WHERE table_name = 'llm_profiles' AND column_name = 'thinking_type'
|
||||
) THEN
|
||||
ALTER TABLE llm_profiles ADD COLUMN thinking_type TEXT NOT NULL DEFAULT '';
|
||||
UPDATE llm_profiles SET thinking_type = 'enabled'
|
||||
WHERE reasoning_effort IN ('low','medium','high','max');
|
||||
UPDATE llm_profiles SET thinking_type = 'disabled', reasoning_effort = ''
|
||||
WHERE reasoning_effort = 'off';
|
||||
END IF;
|
||||
END $$;
|
||||
|
||||
-- LLM 轮询熔断状态：某个配置连续失败(余额不足/key 失效/限流)后进入冷却，冷却期内
|
||||
-- 轮询直接跳过它。内存态为准，这里落库只为重启后不丢冷却窗口——加载时只取尚未
|
||||
-- 到期的行(open_until > now)，已到期的自然回到"正常"，等下一次调用半开试探。
|
||||
|
||||
Reference in new issue
Block a user