fix(vlm): allow configurable temperature for VLM models

Some VLM models (e.g. certain vision models) only accept temperature=1.
Previously temperature was hardcoded to 0.1. Now reads from model's
ExtraConfig["temperature"] with fallback to default 0.1.

Usage: set ExtraConfig.temperature=1 in the VLM model configuration.
This commit is contained in:
lawrence
2026-07-01 11:08:03 +08:00
committed by lyingbug
parent 9682ad67a6
commit fdfc7297de
+22 -9
View File
@@ -40,10 +40,11 @@ func vlmHTTPTimeout() time.Duration {
// RemoteAPIVLM implements VLM via an OpenAI-compatible chat completions API.
type RemoteAPIVLM struct {
modelName string
modelID string
client *openai.Client
baseURL string
modelName string
modelID string
client *openai.Client
baseURL string
temperature float32
}
// NewRemoteAPIVLM creates a remote-API backed VLM instance.
@@ -81,11 +82,23 @@ func NewRemoteAPIVLM(config *Config) (*RemoteAPIVLM, error) {
apiCfg.HTTPClient = httpClient
}
temp := defaultTemp
if config.Extra != nil {
if v, ok := config.Extra["temperature"]; ok {
if vs, ok := v.(string); ok {
if f, err := strconv.ParseFloat(vs, 32); err == nil {
temp = float32(f)
}
}
}
}
return &RemoteAPIVLM{
modelName: config.ModelName,
modelID: config.ModelID,
client: openai.NewClientWithConfig(apiCfg),
baseURL: config.BaseURL,
modelName: config.ModelName,
modelID: config.ModelID,
client: openai.NewClientWithConfig(apiCfg),
baseURL: config.BaseURL,
temperature: temp,
}, nil
}
@@ -124,7 +137,7 @@ func (v *RemoteAPIVLM) Predict(ctx context.Context, imgBytesList [][]byte, promp
},
},
MaxTokens: defaultMaxToks,
Temperature: defaultTemp,
Temperature: v.temperature,
}
totalImageSize := 0