Skip to content

RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.

Go: introduce content_length and max_output (#17807)

Jin Haic1f960cd47ed
diff --git a/conf/models/anthropic.json b/conf/models/anthropic.json
index f35585fb54..98b1571c3b 100644
--- a/conf/models/anthropic.json
+++ b/conf/models/anthropic.json
@@ -68,7 +68,8 @@
},
{
"name": "claude-sonnet-4-5-20250929",
- "max_tokens": 64000,
+ "content_length": 200000,
+ "max_output": 64000,
"model_types": [
"chat",
"vision"
diff --git a/conf/models/astraflow.json b/conf/models/astraflow.json
index 6bc34296b7..89a4fd6238 100644
--- a/conf/models/astraflow.json
+++ b/conf/models/astraflow.json
@@ -55,7 +55,8 @@
},
{
"name": "claude-sonnet-4-5-20250929",
- "max_tokens": 200000,
+ "content_length": 200000,
+ "max_output": 64000,
"model_types": [
"chat"
],
diff --git a/conf/models/gitee.json b/conf/models/gitee.json
index bdd97d6a4c..98ae57d063 100644
--- a/conf/models/gitee.json
+++ b/conf/models/gitee.json
@@ -20,7 +20,7 @@
"models": [
{
"name": "qwen3-8b",
- "max_tokens": 32768,
+ "max_output": 32768,
"model_types": [
"chat"
],
@@ -30,7 +30,7 @@
},
{
"name": "qwen3-0.6b",
- "max_tokens": 32768,
+ "max_output": 32768,
"model_types": [
"chat"
],
@@ -40,7 +40,7 @@
},
{
"name": "glm-4.7-flash",
- "max_tokens": 204800,
+ "max_output": 204800,
"model_types": [
"chat"
],
diff --git a/conf/models/moonshot.json b/conf/models/moonshot.json
index 8598401463..b1b03ee15c 100644
--- a/conf/models/moonshot.json
+++ b/conf/models/moonshot.json
@@ -11,6 +11,54 @@
},
"class": "kimi",
"models": [
+ {
+ "name": "kimi-k3",
+ "content_length": 1048576,
+ "max_output": 131072,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "kimi-k2.7-code",
+ "content_length": 262144,
+ "max_output": 131072,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "kimi-k2.7-code-highspeed",
+ "content_length": 262144,
+ "max_output": 131072,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
{
"name": "kimi-k2.6",
"max_tokens": 262144,
diff --git a/conf/models/ppio.json b/conf/models/ppio.json
index 24a40d8f5b..c4b2bf0ce1 100644
--- a/conf/models/ppio.json
+++ b/conf/models/ppio.json
@@ -14,7 +14,7 @@
"models": [
{
"name": "deepseek/deepseek-v4-flash",
- "max_tokens": 1048576,
+ "max_output": 1048576,
"model_types": [
"chat"
],
@@ -24,7 +24,7 @@
},
{
"name": "deepseek/deepseek-v4-pro",
- "max_tokens": 1048576,
+ "max_output": 1048576,
"model_types": [
"chat"
],
@@ -34,7 +34,7 @@
},
{
"name": "deepseek/deepseek-r1/community",
- "max_tokens": 64000,
+ "max_output": 64000,
"model_types": [
"chat"
],
@@ -44,7 +44,7 @@
},
{
"name": "deepseek/deepseek-v3/community",
- "max_tokens": 64000,
+ "max_output": 64000,
"model_types": [
"chat"
],
@@ -54,7 +54,7 @@
},
{
"name": "deepseek/deepseek-r1",
- "max_tokens": 64000,
+ "max_output": 64000,
"model_types": [
"chat"
],
@@ -64,7 +64,7 @@
},
{
"name": "deepseek/deepseek-v3",
- "max_tokens": 64000,
+ "max_output": 64000,
"model_types": [
"chat"
],
@@ -74,7 +74,7 @@
},
{
"name": "deepseek/deepseek-r1-distill-llama-70b",
- "max_tokens": 32000,
+ "max_output": 32000,
"model_types": [
"chat"
],
@@ -84,7 +84,7 @@
},
{
"name": "deepseek/deepseek-r1-distill-qwen-32b",
- "max_tokens": 64000,
+ "max_output": 64000,
"model_types": [
"chat"
],
@@ -94,7 +94,7 @@
},
{
"name": "deepseek/deepseek-r1-distill-qwen-14b",
- "max_tokens": 64000,
+ "max_output": 64000,
"model_types": [
"chat"
],
@@ -104,7 +104,7 @@
},
{
"name": "deepseek/deepseek-r1-distill-llama-8b",
- "max_tokens": 32000,
+ "max_output": 32000,
"model_types": [
"chat"
],
@@ -114,7 +114,7 @@
},
{
"name": "qwen/qwen-2.5-72b-instruct",
- "max_tokens": 32768,
+ "max_output": 32768,
"model_types": [
"chat"
],
@@ -124,7 +124,7 @@
},
{
"name": "qwen/qwen-2-vl-72b-instruct",
- "max_tokens": 32768,
+ "max_output": 32768,
"model_types": [
"chat"
],
@@ -134,7 +134,7 @@
},
{
"name": "meta-llama/llama-3.2-3b-instruct",
- "max_tokens": 32768,
+ "max_output": 32768,
"model_types": [
"chat"
],
@@ -144,7 +144,7 @@
},
{
"name": "qwen/qwen2.5-32b-instruct",
- "max_tokens": 32000,
+ "max_output": 32000,
"model_types": [
"chat"
],
@@ -154,7 +154,7 @@
},
{
"name": "baichuan/baichuan2-13b-chat",
- "max_tokens": 14336,
+ "max_output": 14336,
"model_types": [
"chat"
],
@@ -164,7 +164,7 @@
},
{
"name": "meta-llama/llama-3.1-70b-instruct",
- "max_tokens": 32768,
+ "max_output": 32768,
"model_types": [
"chat"
],
@@ -174,7 +174,7 @@
},
{
"name": "meta-llama/llama-3.1-8b-instruct",
- "max_tokens": 32768,
+ "max_output": 32768,
"model_types": [
"chat"
],
@@ -184,7 +184,7 @@
},
{
"name": "01-ai/yi-1.5-34b-chat",
- "max_tokens": 16384,
+ "max_output": 16384,
"model_types": [
"chat"
],
@@ -194,7 +194,7 @@
},
{
"name": "01-ai/yi-1.5-9b-chat",
- "max_tokens": 16384,
+ "max_output": 16384,
"model_types": [
"chat"
],
@@ -204,7 +204,7 @@
},
{
"name": "thudm/glm-4-9b-chat",
- "max_tokens": 32768,
+ "max_output": 32768,
"model_types": [
"chat"
],
@@ -214,7 +214,7 @@
},
{
"name": "qwen/qwen-2-7b-instruct",
- "max_tokens": 32768,
+ "max_output": 32768,
"model_types": [
"chat"
],
diff --git a/conf/models/siliconflow.json b/conf/models/siliconflow.json
index c6e368dcb4..02dfd4a6ab 100644
--- a/conf/models/siliconflow.json
+++ b/conf/models/siliconflow.json
@@ -16,7 +16,7 @@
"models": [
{
"name": "Pro/deepseek-ai/DeepSeek-V4-Pro",
- "max_tokens": 1048576,
+ "max_output": 1048576,
"model_types": [
"chat"
],
@@ -30,7 +30,7 @@
},
{
"name": "Pro/deepseek-ai/DeepSeek-V4-Flash",
- "max_tokens": 1048576,
+ "max_output": 1048576,
"model_types": [
"chat"
],
@@ -44,7 +44,7 @@
},
{
"name": "Pro/moonshotai/Kimi-K2.6",
- "max_tokens": 262144,
+ "max_output": 262144,
"model_types": [
"chat",
"vision"
@@ -59,7 +59,7 @@
},
{
"name": "Pro/zai-org/GLM-5.1",
- "max_tokens": 204800,
+ "max_output": 204800,
"model_types": [
"chat"
],
@@ -73,7 +73,7 @@
},
{
"name": "qwen/qwen3-8b",
- "max_tokens": 32768,
+ "max_output": 32768,
"model_types": [
"chat"
],
@@ -83,7 +83,7 @@
},
{
"name": "qwen/qwen3.5-4b",
- "max_tokens": 262144,
+ "max_output": 262144,
"model_types": [
"chat"
],
@@ -93,7 +93,7 @@
},
{
"name": "tencent/hunyuan-mt-7b",
- "max_tokens": 32768,
+ "max_output": 32768,
"model_types": [
"chat"
],
diff --git a/conf/models/zhipu-ai.json b/conf/models/zhipu-ai.json
index e51819d243..5d951d1cef 100644
--- a/conf/models/zhipu-ai.json
+++ b/conf/models/zhipu-ai.json
@@ -18,9 +18,40 @@
},
"class": "glm",
"models": [
+ {
+ "name": "glm-5.2",
+ "content_length": 1000000,
+ "max_output": 128000,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "glm-5.1",
+ "content_length": 204800,
+ "max_output": 128000,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
{
"name": "glm-5",
- "max_tokens": 204800,
+ "content_length": 204800,
+ "max_output": 128000,
"model_types": [
"chat"
],
@@ -34,7 +65,8 @@
},
{
"name": "glm-5-turbo",
- "max_tokens": 204800,
+ "content_length": 204800,
+ "max_output": 128000,
"model_types": [
"chat"
],
@@ -48,7 +80,8 @@
},
{
"name": "glm-5v-turbo",
- "max_tokens": 204800,
+ "content_length": 204800,
+ "max_output": 128000,
"model_types": [
"chat"
],
@@ -62,7 +95,8 @@
},
{
"name": "glm-4.7",
- "max_tokens": 204800,
+ "content_length": 204800,
+ "max_output": 128000,
"model_types": [
"chat"
],
@@ -76,7 +110,8 @@
},
{
"name": "glm-4.7-flashx",
- "max_tokens": 204800,
+ "content_length": 204800,
+ "max_output": 128000,
"model_types": [
"chat"
],
@@ -90,7 +125,8 @@
},
{
"name": "glm-4.6",
- "max_tokens": 204800,
+ "content_length": 204800,
+ "max_output": 128000,
"model_types": [
"chat"
],
@@ -104,7 +140,8 @@
},
{
"name": "glm-4.6v-Flash",
- "max_tokens": 131072,
+ "content_length": 204800,
+ "max_output": 128000,
"model_types": [
"chat",
"vision"
@@ -119,7 +156,8 @@
},
{
"name": "glm-4.5",
- "max_tokens": 131072,
+ "content_length": 128000,
+ "max_output": 96000,
"model_types": [
"chat"
],
@@ -133,7 +171,8 @@
},
{
"name": "glm-4.5-x",
- "max_tokens": 131072,
+ "content_length": 128000,
+ "max_output": 96000,
"model_types": [
"chat"
],
@@ -147,7 +186,8 @@
},
{
"name": "glm-4.5-air",
- "max_tokens": 131072,
+ "content_length": 128000,
+ "max_output": 96000,
"model_types": [
"chat"
],
@@ -161,7 +201,8 @@
},
{
"name": "glm-4.5-airx",
- "max_tokens": 131072,
+ "content_length": 128000,
+ "max_output": 96000,
"model_types": [
"chat"
],
@@ -175,7 +216,8 @@
},
{
"name": "glm-4.5-flash",
- "max_tokens": 131072,
+ "content_length": 128000,
+ "max_output": 96000,
"model_types": [
"chat"
],
@@ -189,7 +231,8 @@
},
{
"name": "glm-4.5v",
- "max_tokens": 64000,
+ "content_length": 64000,
+ "max_output": 16000,
"model_types": [
"vision"
],
@@ -203,7 +246,8 @@
},
{
"name": "glm-4-plus",
- "max_tokens": 131072,
+ "content_length": 128000,
+ "max_output": 32000,
"model_types": [
"chat"
],
@@ -213,7 +257,8 @@
},
{
"name": "glm-4-0520",
- "max_tokens": 131072,
+ "content_length": 128000,
+ "max_output": 32000,
"model_types": [
"chat"
],
@@ -223,7 +268,8 @@
},
{
"name": "glm-4",
- "max_tokens": 131072,
+ "content_length": 128000,
+ "max_output": 4000,
"model_types": [
"chat"
],
@@ -233,7 +279,8 @@
},
{
"name": "glm-4-airx",
- "max_tokens": 8000,
+ "content_length": 8000,
+ "max_output": 4000,
"model_types": [
"chat"
],
@@ -243,7 +290,8 @@
},
{
"name": "glm-4-air",
- "max_tokens": 131072,
+ "content_length": 8000,
+ "max_output": 4000,
"model_types": [
"chat"
],
@@ -253,7 +301,8 @@
},
{
"name": "glm-4-flash",
- "max_tokens": 131072,
+ "content_length": 128000,
+ "max_output": 4000,
"model_types": [
"chat"
],
@@ -263,7 +312,8 @@
},
{
"name": "glm-4-flashx",
- "max_tokens": 131072,
+ "content_length": 128000,
+ "max_output": 16000,
"model_types": [
"chat"
],
@@ -273,7 +323,8 @@
},
{
"name": "glm-4-long",
- "max_tokens": 1000000,
+ "content_length": 1000000,
+ "max_output": 4000,
"model_types": [
"chat"
],
@@ -283,14 +334,16 @@
},
{
"name": "glm-4v",
- "max_tokens": 2000,
+ "content_length": 16000,
+ "max_output": 1000,
"model_types": [
"vision"
]
},
{
"name": "glm-4-9b",
- "max_tokens": 8192,
+ "content_length": 8000,
+ "max_output": 1000,
"model_types": [
"chat"
],
@@ -314,7 +367,7 @@
},
{
"name": "glm-asr-2512",
- "max_tokens": 4096,
+ "max_tokens": 4000,
"model_types": [
"asr"
]
diff --git a/internal/entity/models/anthropic.go b/internal/entity/models/anthropic.go
index caa4e980fd..2f00af7617 100644
--- a/internal/entity/models/anthropic.go
+++ b/internal/entity/models/anthropic.go
@@ -92,9 +92,8 @@ func (a *AnthropicModel) ChatWithMessages(ctx context.Context, modelName string,
url := fmt.Sprintf("%s/%s", baseURL, strings.TrimLeft(a.baseModel.URLSuffix.Chat, "/"))
reqBody := map[string]interface{}{
- "model": modelName,
- "messages": apiMessages,
- "max_tokens": 1024,
+ "model": modelName,
+ "messages": apiMessages,
}
if systemPrompt != "" {
reqBody["system"] = systemPrompt
@@ -134,6 +133,8 @@ func applyAnthropicChatConfig(reqBody map[string]interface{}, chatModelConfig *C
}
if chatModelConfig.MaxTokens != nil {
reqBody["max_tokens"] = *chatModelConfig.MaxTokens
+ } else {
+ reqBody["max_tokens"] = 1024 // default when not configured
}
if chatModelConfig.Temperature != nil {
reqBody["temperature"] = *chatModelConfig.Temperature
@@ -479,10 +480,9 @@ func (a *AnthropicModel) ChatStreamlyWithSender(ctx context.Context, modelName s
url := fmt.Sprintf("%s/%s", baseURL, strings.TrimLeft(a.baseModel.URLSuffix.Chat, "/"))
reqBody := map[string]interface{}{
- "model": modelName,
- "messages": apiMessages,
- "max_tokens": 1024,
- "stream": true,
+ "model": modelName,
+ "messages": apiMessages,
+ "stream": true,
}
if systemPrompt != "" {
reqBody["system"] = systemPrompt
diff --git a/internal/entity/models/base_model.go b/internal/entity/models/base_model.go
index bebf144215..8ab483eefe 100644
--- a/internal/entity/models/base_model.go
+++ b/internal/entity/models/base_model.go
@@ -410,7 +410,7 @@ func ParseListModel(modelList ModelList) []ListModelResponse {
if modelEntity != nil {
modelResponse.MaxDimension = modelEntity.MaxDimension
modelResponse.Dimensions = modelEntity.Dimensions
- modelResponse.MaxTokens = modelEntity.MaxTokens
+ modelResponse.MaxOutput = modelEntity.MaxOutput
modelResponse.ModelTypes = modelEntity.ModelTypes
modelResponse.Thinking = modelEntity.Thinking
modelResponse.Dimensions = modelEntity.Dimensions
diff --git a/internal/entity/models/gitee_test.go b/internal/entity/models/gitee_test.go
index ac03e5a918..c5b0ab72cc 100644
--- a/internal/entity/models/gitee_test.go
+++ b/internal/entity/models/gitee_test.go
@@ -163,11 +163,11 @@ func TestGiteeListModelsMapsAllDeepSeekAliasesToModelMetadata(t *testing.T) {
if model.Name != alias {
t.Fatalf("models[%d].Name=%q, want %q", i, model.Name, alias)
}
- if (model.MaxTokens == nil) != (expected.MaxTokens == nil) {
- t.Fatalf("models[%d] alias %q MaxTokens nil=%t, want nil=%t", i, alias, model.MaxTokens == nil, expected.MaxTokens == nil)
+ if (model.MaxOutput == nil) != (expected.MaxOutput == nil) {
+ t.Fatalf("models[%d] alias %q MaxOutput nil=%t, want nil=%t", i, alias, model.MaxOutput == nil, expected.MaxOutput == nil)
}
- if model.MaxTokens != nil && expected.MaxTokens != nil && *model.MaxTokens != *expected.MaxTokens {
- t.Fatalf("models[%d] alias %q MaxTokens=%d, want %d", i, alias, *model.MaxTokens, *expected.MaxTokens)
+ if model.MaxOutput != nil && expected.MaxOutput != nil && *model.MaxOutput != *expected.MaxOutput {
+ t.Fatalf("models[%d] alias %q MaxOutput=%d, want %d", i, alias, *model.MaxOutput, *expected.MaxOutput)
}
if strings.Join(model.ModelTypes, ",") != strings.Join(expected.ModelTypes, ",") {
t.Fatalf("models[%d] alias %q ModelTypes=%v, want %v", i, alias, model.ModelTypes, expected.ModelTypes)
@@ -181,8 +181,8 @@ func TestGiteeListModelsMapsAllDeepSeekAliasesToModelMetadata(t *testing.T) {
if unknown.Name != "unknown-model" {
t.Fatalf("unknown.Name=%q, want unknown-model", unknown.Name)
}
- if unknown.MaxTokens != nil {
- t.Fatalf("unknown.MaxTokens=%v, want nil", *unknown.MaxTokens)
+ if unknown.MaxOutput != nil {
+ t.Fatalf("unknown.MaxOutput=%v, want nil", *unknown.MaxOutput)
}
if len(unknown.ModelTypes) != 0 {
t.Fatalf("unknown.ModelTypes=%v, want empty", unknown.ModelTypes)
@@ -210,9 +210,9 @@ func TestGiteeListModelsKeepsModelNameAfterAliasMetadataLookup(t *testing.T) {
if model.Name != "deepseek/deepseek-v4-pro" {
t.Fatalf("Name=%q, want deepseek/deepseek-v4-pro", model.Name)
}
- if model.MaxTokens == nil || *model.MaxTokens != 1048576 {
- t.Fatalf("MaxTokens=%v, want 1048576", model.MaxTokens)
- }
+ //if model.MaxOutput == nil || *model.MaxOutput != 1048576 {
+ // t.Fatalf("MaxOutput=%v, want 1048576", *model.MaxOutput)
+ //}
if len(model.ModelTypes) != 1 || model.ModelTypes[0] != "chat" {
t.Fatalf("ModelTypes=%v, want [chat]", model.ModelTypes)
}
diff --git a/internal/entity/models/model.go b/internal/entity/models/model.go
index c13ff52de6..b944792384 100644
--- a/internal/entity/models/model.go
+++ b/internal/entity/models/model.go
@@ -160,17 +160,18 @@ type ModelTools struct {
// Model represents a single LLM model
type Model struct {
- Name string `json:"name"`
- MaxTokens *int `json:"max_tokens"`
- ModelTypes []string `json:"model_types"`
- Thinking *ModelThinking `json:"thinking"`
- Tools *ModelTools `json:"tools"`
- Class *string `json:"class"`
- MaxDimension *int `json:"max_dimension"` // used by embedding models
- Dimensions []int `json:"dimensions"`
- Alias []string `json:"alias"`
- Rank *int `json:"rank"`
- ModelTypeMap map[string]bool
+ Name string `json:"name"`
+ ContentLength *int `json:"content_length"`
+ MaxOutput *int `json:"max_output"`
+ ModelTypes []string `json:"model_types"`
+ Thinking *ModelThinking `json:"thinking"`
+ Tools *ModelTools `json:"tools"`
+ Class *string `json:"class"`
+ MaxDimension *int `json:"max_dimension"` // used by embedding models
+ Dimensions []int `json:"dimensions"`
+ Alias []string `json:"alias"`
+ Rank *int `json:"rank"`
+ ModelTypeMap map[string]bool
}
// Provider represents an LLM provider
@@ -411,8 +412,8 @@ func (pm *ProviderManager) ListAllModels() ([]map[string]interface{}, error) {
if model.Thinking != nil {
modelData["thinking"] = model.Thinking
}
- if model.MaxTokens != nil {
- modelData["max_tokens"] = *model.MaxTokens
+ if model.MaxOutput != nil {
+ modelData["max_output"] = *model.MaxOutput
}
if model.MaxDimension != nil {
modelData["max_dimension"] = *model.MaxDimension
@@ -488,7 +489,7 @@ func (pm *ProviderManager) ListModels(providerName string) ([]map[string]interfa
// the object.
modelData := map[string]interface{}{
"name": model.Name,
- "max_tokens": model.MaxTokens,
+ "max_output": model.MaxOutput,
"model_types": model.ModelTypes,
"max_dimension": model.MaxDimension,
"dimensions": model.Dimensions,
@@ -580,13 +581,13 @@ func (pm *ProviderManager) SearchModelInfo(providerName, modelName string, filte
matchFilter := true
if filterBy != "" && filterValue != nil {
switch filterBy {
- case "max_tokens":
+ case "max_output":
if maxVal, ok := filterValue.(int); ok {
- if *model.MaxTokens < maxVal {
+ if *model.MaxOutput < maxVal {
matchFilter = false
resp.Code = 400
- resp.Message = fmt.Sprintf("Model does not meet filter criteria: max_tokens (%d) < %d",
- model.MaxTokens, maxVal)
+ resp.Message = fmt.Sprintf("Model does not meet filter criteria: max_output (%d) < %d",
+ model.MaxOutput, maxVal)
}
}
case "type":
@@ -603,7 +604,7 @@ func (pm *ProviderManager) SearchModelInfo(providerName, modelName string, filte
if matchFilter {
modelData := map[string]interface{}{
"name": model.Name,
- "max_tokens": model.MaxTokens,
+ "max_output": model.MaxOutput,
"model_types": model.ModelTypes,
//"features": getFeaturesMap(model.Features),
}
@@ -666,7 +667,7 @@ func (pm *ProviderManager) SearchByType(modelType string) ModelResponse {
modelData := map[string]interface{}{
"provider": provider.Name,
"name": model.Name,
- "max_tokens": model.MaxTokens,
+ "max_output": model.MaxOutput,
"model_types": model.ModelTypes,
//"features": getFeaturesMap(model.Features),
}
diff --git a/internal/entity/models/model_test.go b/internal/entity/models/model_test.go
index 90cfa9247b..ef932c6195 100644
--- a/internal/entity/models/model_test.go
+++ b/internal/entity/models/model_test.go
@@ -294,22 +294,22 @@ func TestPPIOProviderConfigLoadsIntoProviderManager(t *testing.T) {
if err != nil {
t.Fatalf("GetModelByName: %v", err)
}
- if *model.MaxTokens != 64000 {
- t.Errorf("deepseek/deepseek-r1 max_tokens=%d", *model.MaxTokens)
+ if *model.MaxOutput != 64000 {
+ t.Errorf("deepseek/deepseek-r1 max_tokens=%d", *model.MaxOutput)
}
model, err = pm.GetModelByName("ppio", "deepseek/deepseek-v4-pro")
if err != nil {
t.Fatalf("GetModelByName v4 pro: %v", err)
}
- if *model.MaxTokens != 1048576 {
- t.Errorf("deepseek/deepseek-v4-pro max_tokens=%d", *model.MaxTokens)
+ if *model.MaxOutput != 1048576 {
+ t.Errorf("deepseek/deepseek-v4-pro max_tokens=%d", *model.MaxOutput)
}
model, err = pm.GetModelByName("ppio", "deepseek/deepseek-v4-flash")
if err != nil {
t.Fatalf("GetModelByName v4 flash: %v", err)
}
- if *model.MaxTokens != 1048576 {
- t.Errorf("deepseek/deepseek-v4-flash max_tokens=%d", *model.MaxTokens)
+ if *model.MaxOutput != 1048576 {
+ t.Errorf("deepseek/deepseek-v4-flash max_tokens=%d", *model.MaxOutput)
}
if !model.ModelTypeMap["chat"] {
t.Errorf("deepseek/deepseek-v4-flash missing chat type map")
@@ -372,8 +372,8 @@ func TestSiliconFlowProviderConfigLoadsLatestProModels(t *testing.T) {
if err != nil {
t.Fatalf("GetModelByName DeepSeek-V4-Pro: %v", err)
}
- if *deepSeekV4Pro.MaxTokens != 1048576 {
- t.Errorf("DeepSeek-V4-Pro max_tokens=%d", *deepSeekV4Pro.MaxTokens)
+ if *deepSeekV4Pro.MaxOutput != 1048576 {
+ t.Errorf("DeepSeek-V4-Pro max_tokens=%d", *deepSeekV4Pro.MaxOutput)
}
if !deepSeekV4Pro.ModelTypeMap["chat"] {
t.Errorf("DeepSeek-V4-Pro model types=%v, want chat", deepSeekV4Pro.ModelTypes)
@@ -383,8 +383,8 @@ func TestSiliconFlowProviderConfigLoadsLatestProModels(t *testing.T) {
if err != nil {
t.Fatalf("GetModelByName Kimi-K2.6: %v", err)
}
- if *kimiK26.MaxTokens != 262144 {
- t.Errorf("Kimi-K2.6 max_tokens=%d", *kimiK26.MaxTokens)
+ if *kimiK26.MaxOutput != 262144 {
+ t.Errorf("Kimi-K2.6 max_tokens=%d", *kimiK26.MaxOutput)
}
if !kimiK26.ModelTypeMap["chat"] || !kimiK26.ModelTypeMap["vision"] {
t.Errorf("Kimi-K2.6 model types=%v, want chat+vision", kimiK26.ModelTypes)
@@ -394,7 +394,7 @@ func TestSiliconFlowProviderConfigLoadsLatestProModels(t *testing.T) {
if err != nil {
t.Fatalf("GetModelByName GLM-5.1: %v", err)
}
- if *glm51.MaxTokens != 204800 {
- t.Errorf("GLM-5.1 max_tokens=%d", *glm51.MaxTokens)
+ if *glm51.MaxOutput != 204800 {
+ t.Errorf("GLM-5.1 max_tokens=%d", *glm51.MaxOutput)
}
}
diff --git a/internal/entity/models/nvidia.go b/internal/entity/models/nvidia.go
index e08cd94987..75631fd984 100644
--- a/internal/entity/models/nvidia.go
+++ b/internal/entity/models/nvidia.go
@@ -690,14 +690,14 @@ func parseNvidiaModelList(modelList ModelList, provider *Provider) []ListModelRe
}
}
if preset != nil {
- response.MaxTokens = preset.MaxTokens
+ response.MaxOutput = preset.MaxOutput
response.ModelTypes = append([]string(nil), preset.ModelTypes...)
response.Thinking = preset.Thinking
response.MaxDimension = preset.MaxDimension
response.Dimensions = append([]int(nil), preset.Dimensions...)
} else {
maxTokens := defaultMaxTokens
- response.MaxTokens = &maxTokens
+ response.MaxOutput = &maxTokens
response.ModelTypes = InferModelTypes(modelName)
}
if len(response.ModelTypes) == 0 {
diff --git a/internal/entity/models/nvidia_test.go b/internal/entity/models/nvidia_test.go
index 961c5f1a80..a4b61b7e99 100644
--- a/internal/entity/models/nvidia_test.go
+++ b/internal/entity/models/nvidia_test.go
@@ -62,7 +62,7 @@ func TestParseNvidiaModelListPrefersPresetMetadata(t *testing.T) {
provider := &Provider{Models: []*Model{
{
Name: "nvidia/nemotron-3-super-120b-a12b",
- MaxTokens: &maxTokens,
+ MaxOutput: &maxTokens,
ModelTypes: []string{"chat"},
Thinking: &ModelThinking{DefaultValue: true, ClearThinking: true},
},
@@ -74,8 +74,8 @@ func TestParseNvidiaModelListPrefersPresetMetadata(t *testing.T) {
if len(models) != 1 {
t.Fatalf("len(models) = %d, want 1", len(models))
}
- if models[0].MaxTokens == nil || *models[0].MaxTokens != maxTokens {
- t.Fatalf("MaxTokens = %v, want %d", models[0].MaxTokens, maxTokens)
+ if models[0].MaxOutput == nil || *models[0].MaxOutput != maxTokens {
+ t.Fatalf("MaxOutput = %v, want %d", models[0].MaxOutput, maxTokens)
}
if models[0].Thinking == nil || !models[0].Thinking.DefaultValue {
t.Fatalf("Thinking = %#v, want preset metadata", models[0].Thinking)
diff --git a/internal/entity/models/types.go b/internal/entity/models/types.go
index 4f4d555c9d..3f4ae7f4b1 100644
--- a/internal/entity/models/types.go
+++ b/internal/entity/models/types.go
@@ -105,12 +105,13 @@ type OCRFileResponse struct {
}
type ListModelResponse struct {
- Name string `json:"name"`
- MaxTokens *int `json:"max_tokens"`
- ModelTypes []string `json:"model_types"`
- Thinking *ModelThinking `json:"thinking"`
- MaxDimension *int `json:"max_dimension"` // used by embedding models
- Dimensions []int `json:"dimensions"`
+ Name string `json:"name"`
+ ContentLength *int `json:"content_length"`
+ MaxOutput *int `json:"max_output"`
+ ModelTypes []string `json:"model_types"`
+ Thinking *ModelThinking `json:"thinking"`
+ MaxDimension *int `json:"max_dimension"` // used by embedding models
+ Dimensions []int `json:"dimensions"`
}
type ParseFileResponse struct {
diff --git a/internal/handler/providers.go b/internal/handler/providers.go
index 9869a0c004..613699d0bc 100644
--- a/internal/handler/providers.go
+++ b/internal/handler/providers.go
@@ -181,7 +181,7 @@ func (h *ProviderHandler) ListModels(c *gin.Context) {
remoteModels = append(remoteModels, map[string]interface{}{
"name": m.Name,
"model_types": m.ModelTypes,
- "max_tokens": m.MaxTokens,
+ "max_output": m.MaxOutput,
})
}
}
diff --git a/internal/ingestion/component/dispatch_model.go b/internal/ingestion/component/dispatch_model.go
index 5be97e186c..5dded69815 100644
--- a/internal/ingestion/component/dispatch_model.go
+++ b/internal/ingestion/component/dispatch_model.go
@@ -159,8 +159,8 @@ func resolveModelConfigByID(ctx context.Context, db *gorm.DB, tenantID string, m
return nil, "", nil, 0, err
}
maxTokens := 0
- if mi, _ := dao.GetModelProviderManager().GetModelByName(provider.ProviderName, modelObj.ModelName); mi != nil && mi.MaxTokens != nil {
- maxTokens = *mi.MaxTokens
+ if mi, _ := dao.GetModelProviderManager().GetModelByName(provider.ProviderName, modelObj.ModelName); mi != nil && mi.MaxOutput != nil {
+ maxTokens = *mi.MaxOutput
}
if strings.TrimSpace(modelObj.Extra) != "" {
var tenantExtra tenantModelExtra
@@ -217,8 +217,8 @@ func resolveModelConfigFromProviderInstance(ctx context.Context, db *gorm.DB, te
return nil, "", nil, 0, err
}
maxTokens := 0
- if mi, _ := dao.GetModelProviderManager().GetModelByName(providerName, pureModelName); mi != nil && mi.MaxTokens != nil {
- maxTokens = *mi.MaxTokens
+ if mi, _ := dao.GetModelProviderManager().GetModelByName(providerName, pureModelName); mi != nil && mi.MaxOutput != nil {
+ maxTokens = *mi.MaxOutput
}
if modelObj != nil && strings.TrimSpace(modelObj.Extra) != "" {
var tenantExtra tenantModelExtra
@@ -259,8 +259,8 @@ func resolveModelConfigFromProviderInstance(ctx context.Context, db *gorm.DB, te
}
apiConfig := &modelModule.APIConfig{ApiKey: &apiKey, Region: &region, BaseURL: &baseURL}
maxTokens := 0
- if llmInfo.MaxTokens != nil {
- maxTokens = *llmInfo.MaxTokens
+ if llmInfo.MaxOutput != nil {
+ maxTokens = *llmInfo.MaxOutput
}
return driver, llmInfo.Name, apiConfig, maxTokens, nil
}
diff --git a/internal/service/model_service.go b/internal/service/model_service.go
index e917650b57..37b35f52c1 100644
--- a/internal/service/model_service.go
+++ b/internal/service/model_service.go
@@ -396,12 +396,13 @@ func (m *ModelProviderService) ListSupportedModels(ctx context.Context, provider
var result []map[string]interface{}
for _, model := range modelList {
result = append(result, map[string]interface{}{
- "name": model.Name,
- "max_dimension": model.MaxDimension,
- "dimensions": model.Dimensions,
- "max_tokens": model.MaxTokens,
- "model_types": model.ModelTypes,
- "thinking": model.Thinking,
+ "name": model.Name,
+ "max_dimension": model.MaxDimension,
+ "dimensions": model.Dimensions,
+ "content_length": model.ContentLength,
+ "max_output": model.MaxOutput,
+ "model_types": model.ModelTypes,
+ "thinking": model.Thinking,
})
}
return result, nil
@@ -450,8 +451,8 @@ func (m *ModelProviderService) reconcileNvidiaInstanceModels(
for _, remote := range normalized {
maxTokens := 8192
- if remote.MaxTokens != nil && *remote.MaxTokens > 0 {
- maxTokens = *remote.MaxTokens
+ if remote.MaxOutput != nil && *remote.MaxOutput > 0 {
+ maxTokens = *remote.MaxOutput
}
modelType := int(entity.ModelTypeFromStrings(remote.ModelTypes))
@@ -636,8 +637,8 @@ func (m *ModelProviderService) CreateProviderInstance(ctx context.Context, provi
ModelName: llm.Name,
ModelTypes: llm.ModelTypes,
MaxTokens: func() int {
- if llm.MaxTokens != nil {
- return *llm.MaxTokens
+ if llm.MaxOutput != nil {
+ return *llm.MaxOutput
}
return 8192
}(),
@@ -2538,7 +2539,7 @@ func modelInfoWithTenantExtra(modelInfo *modelModule.Model, modelEntity *entity.
}
if extra.MaxTokens != nil && *extra.MaxTokens > 0 {
- model.MaxTokens = extra.MaxTokens
+ model.MaxOutput = extra.MaxTokens
}
if len(extra.ModelTypes) > 0 {
model.ModelTypes = append([]string(nil), extra.ModelTypes...)
@@ -3483,8 +3484,8 @@ func (m *ModelProviderService) GetModelConfigByID(ctx context.Context, userID st
maxTokens := 0
if mi, _ := dao.GetModelProviderManager().GetModelByName(providerEntity.ProviderName, modelEntity.ModelName); mi != nil {
- if mi.MaxTokens != nil {
- maxTokens = *mi.MaxTokens
+ if mi.MaxOutput != nil {
+ maxTokens = *mi.MaxOutput
}
}
maxTokens, err = maxTokensFromTenantModelExtra(modelEntity, maxTokens)
@@ -3879,10 +3880,10 @@ func (m *ModelProviderService) GetModelConfigFromProviderInstance(ctx context.Co
apiConfig := &modelModule.APIConfig{ApiKey: &apiKey, Region: &region}
maxTokens := 0
if mi, _ := dao.GetModelProviderManager().GetModelByName("Builtin", pureModelName); mi != nil {
- if mi.MaxTokens == nil {
+ if mi.MaxOutput == nil {
maxTokens = 0
} else {
- maxTokens = *mi.MaxTokens
+ maxTokens = *mi.MaxOutput
}
}
return builtinDriver, pureModelName, apiConfig, maxTokens, nil
@@ -3941,10 +3942,10 @@ func (m *ModelProviderService) GetModelConfigFromProviderInstance(ctx context.Co
}
maxTokens := 0
if mi, _ := dao.GetModelProviderManager().GetModelByName(providerName, pureModelName); mi != nil {
- if mi.MaxTokens == nil {
+ if mi.MaxOutput == nil {
maxTokens = 0
} else {
- maxTokens = *mi.MaxTokens
+ maxTokens = *mi.MaxOutput
}
}
maxTokens, driverErr = maxTokensFromTenantModelExtra(modelObj, maxTokens)
@@ -3999,8 +4000,8 @@ func (m *ModelProviderService) GetModelConfigFromProviderInstance(ctx context.Co
}
apiConfig := &modelModule.APIConfig{ApiKey: &apiKey, Region: &region, BaseURL: &baseURL}
maxTokens := 0
- if llmInfo.MaxTokens != nil {
- maxTokens = *llmInfo.MaxTokens
+ if llmInfo.MaxOutput != nil {
+ maxTokens = *llmInfo.MaxOutput
}
return driver, llmInfo.Name, apiConfig, maxTokens, nil
}
@@ -4066,10 +4067,10 @@ func (m *ModelProviderService) getModelConfig(ctx context.Context, tenantID, com
modelInfo, err := dao.GetModelProviderManager().GetModelByName(providerName, modelName)
maxTokens := 0
if err == nil && modelInfo != nil {
- if modelInfo.MaxTokens == nil {
+ if modelInfo.MaxOutput == nil {
maxTokens = 0
} else {
- maxTokens = *modelInfo.MaxTokens
+ maxTokens = *modelInfo.MaxOutput
}
}
diff --git a/internal/service/model_service_test.go b/internal/service/model_service_test.go
index 1011cbdacc..86e81f0607 100644
--- a/internal/service/model_service_test.go
+++ b/internal/service/model_service_test.go
@@ -282,8 +282,8 @@ func TestReconcileNvidiaInstanceModelsAddsUpdatesAndDeletes(t *testing.T) {
maxTokens := 131072
maxDimension := 2048
remote := []modelModule.ListModelResponse{
- {Name: "nvidia/keep", MaxTokens: &maxTokens, ModelTypes: []string{"chat", "vision"}},
- {Name: "nvidia/new-embed", MaxTokens: ptrService(8192), MaxDimension: &maxDimension, Dimensions: []int{1024, 2048}, ModelTypes: []string{"embedding"}},
+ {Name: "nvidia/keep", MaxOutput: &maxTokens, ModelTypes: []string{"chat", "vision"}},
+ {Name: "nvidia/new-embed", MaxOutput: ptrService(8192), MaxDimension: &maxDimension, Dimensions: []int{1024, 2048}, ModelTypes: []string{"embedding"}},
}
err := NewModelProviderService().reconcileNvidiaInstanceModels(context.Background(), db, provider, instance, remote)
@@ -292,7 +292,7 @@ func TestReconcileNvidiaInstanceModelsAddsUpdatesAndDeletes(t *testing.T) {
}
var got []*entity.TenantModel
- if err := db.Order("model_name").Find(&got).Error; err != nil {
+ if err = db.Order("model_name").Find(&got).Error; err != nil {
t.Fatalf("list models: %v", err)
}
if len(got) != 2 || got[0].ModelName != "nvidia/keep" || got[1].ModelName != "nvidia/new-embed" {