tt/ttstd-infinite-canvas
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

merge: official v1.5.6 with multimodal storyboard enhancements

Browse Source
This commit is contained in:
Codex
2026-09-20 13:05:42 +08:00
parent 6d355380cb 473d535ec1
commit d87c3944b7
199 changed files with 11465 additions and 2971 deletions
Download Patch File Download Diff File Expand all files Collapse all files

33
backend/internal/app/model_capability.go
View File

@@ -23,8 +23,13 @@ type ModelCapabilityConfig struct {
type TextCapabilityConfig struct {
// Streaming controls whether this model accepts upstream SSE text responses.
// A nil value is treated as true for backwards compatibility with older configs.
Streaming *bool `json:"streaming,omitempty"`
References TextReferenceConfig `json:"references"`
Streaming *bool `json:"streaming,omitempty"`
// ContextWindowTokens is the provider's total input plus output context window.
// It is a model contract, not an application transport ceiling.
ContextWindowTokens int `json:"contextWindowTokens"`
// MaxOutputTokens is the provider's maximum completion/reasoning budget.
MaxOutputTokens int `json:"maxOutputTokens"`
References TextReferenceConfig `json:"references"`
}
type TextReferenceConfig struct {
@@ -212,8 +217,10 @@ func DefaultModelCapabilityConfigForModel(protocol string, modelName string) *Mo
if isKnownMultimodalTextModel(protocol, modelName) {
textReferences.MaxImages = 16
textReferences.MaxImageBytes = 30 * 1024 * 1024
textReferences.MaxVideos = 3
textReferences.MaxVideoBytes = 200 * 1024 * 1024
}
text := &TextCapabilityConfig{Streaming: &streaming, References: textReferences}
text := &TextCapabilityConfig{Streaming: &streaming, ContextWindowTokens: 128000, MaxOutputTokens: 16384, References: textReferences}
video := &VideoCapabilityConfig{
References: VideoReferenceConfig{PromptMaxChars: DefaultVideoPromptMaxChars, MinImages: 0, MaxImages: 9, MaxImageBytes: 30 * 1024 * 1024, MaxVideos: 0, MaxVideoBytes: 0, MaxVideoDuration: 0, MaxAudios: 0, MaxAudioBytes: 0, MaxAudioDuration: 0},
Duration: VideoDurationConfig{Selection: "range", Min: 1, Max: 15, Step: 1, Default: 6},
@@ -333,17 +340,27 @@ func NormalizeModelCapabilityConfigForModel(capability string, protocol string,
}
text := *input.Text
if isKnownMultimodalTextModel(protocol, modelName) && text.References.MaxImages == 0 {
// 旧配置是在文本模型默认不接收图片时生成的；Gemini 文本模型
// 通过兼容协议仍然会接收图片 parts，因此自动修复旧的 0 上限。
text.References.MaxImages = 16
if text.References.MaxImageBytes == 0 {
text.References.MaxImageBytes = 30 * 1024 * 1024
}
}
if isKnownMultimodalTextModel(protocol, modelName) && text.References.MaxVideos == 0 {
text.References.MaxVideos = 3
if text.References.MaxVideoBytes == 0 {
text.References.MaxVideoBytes = 200 * 1024 * 1024
}
}
if text.Streaming == nil {
streaming := true
text.Streaming = &streaming
}
if text.ContextWindowTokens == 0 {
text.ContextWindowTokens = 128000
}
if text.MaxOutputTokens == 0 {
text.MaxOutputTokens = 16384
}
value := &ModelCapabilityConfig{Version: 1, Text: &text}
if err := validateTextCapabilityConfig(value.Text); err != nil {
return nil, err
@@ -551,6 +568,12 @@ func addInputConstraint(inputs map[string]InputConstraint, name string, min int,
}
func validateTextCapabilityConfig(value *TextCapabilityConfig) error {
if value.ContextWindowTokens < 4096 || value.ContextWindowTokens > 10000000 {
return BadAuthRequest("文本模型上下文窗口必须在 4096-10000000 Token 之间")
}
if value.MaxOutputTokens < 256 || value.MaxOutputTokens > 1000000 || value.MaxOutputTokens >= value.ContextWindowTokens {
return BadAuthRequest("文本模型最大输出 Token 必须小于上下文窗口且在 256-1000000 之间")
}
if value.References.PromptMaxChars < 1 || value.References.PromptMaxChars > 1000000 {
return BadAuthRequest("提示词最大字符数必须在 1-1000000 之间")
}