Skip to content

fix: allow warm up with system prompt #632

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Oct 4, 2025
+40 −15
Merged

fix: allow warm up with system prompt #632

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
8a88903
fix(run): pass param to serve
RemiliaForever Oct 3, 2025
cc249f2
fix(run): add system prompt in warm up request
RemiliaForever Oct 3, 2025
d61ef79
fix(serve): fix typo
RemiliaForever Oct 3, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 14 additions & 2 deletions runner/cmd/nexa-cli/run.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,6 +37,7 @@ func run() *cobra.Command {

llmFlags := pflag.NewFlagSet("LLM/VLM Model", pflag.ExitOnError)
llmFlags.SortFlags = false
llmFlags.BoolVarP(&enableThink, "think", "", true, "enable thinking mode")
llmFlags.StringVarP(&systemPrompt, "system-prompt", "s", "", "system prompt to set model behavior")
runCmd.Flags().AddFlagSet(llmFlags)

Expand Down Expand Up @@ -186,10 +187,14 @@ func runFunc(cmd *cobra.Command, args []string) {
// warm up
spin := render.NewSpinner("loading model...")
spin.Start()
_, err = client.Chat.Completions.New(context.TODO(), openai.ChatCompletionNewParams{
warmUpRequest := openai.ChatCompletionNewParams{
Messages: nil,
Model: model,
})
}
if systemPrompt != "" {
warmUpRequest.Messages = append(warmUpRequest.Messages, openai.SystemMessage(systemPrompt))
}
_, err = client.Chat.Completions.New(context.TODO(), warmUpRequest)
spin.Stop()

if err != nil {
Expand Down Expand Up @@ -257,6 +262,13 @@ func runFunc(cmd *cobra.Command, args []string) {
FrequencyPenalty: openai.Float(float64(frequencyPenalty)),
Seed: openai.Int(int64(seed)),
},

option.WithJSONSet("enable_think", enableThink),
option.WithJSONSet("top_k", topK),
option.WithJSONSet("min_p", minP),
option.WithJSONSet("repetition_penalty", repetitionPenalty),
option.WithJSONSet("grammar_path", grammarPath),
option.WithJSONSet("grammar_string", grammarString),
option.WithJSONSet("enable_json", enableJson),
option.WithHeaderAdd("Nexa-KeepCache", "true"))

Expand Down
39 changes: 26 additions & 13 deletions runner/server/handler/completion.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,20 +29,33 @@ type ChatCompletionNewParams openai.ChatCompletionNewParams
// ChatCompletionRequest defines the request body for the chat completions API.
// example: { "model": "nexaml/nexaml-models", "messages": [ { "role": "user", "content": "why is the sky blue?" } ] }
type ChatCompletionRequest struct {
Stream bool `json:"stream" default:"false"`
Stream bool `json:"stream"`

EnableThink bool `json:"enable_think" default:"true"`

TopK int32 `json:"top_k" default:"0"`
MinP float32 `json:"min_p" default:"0.0"`
ReqetitionPenalty float32 `json:"repetition_penalty" default:"1.0"`
GrammarPath string `json:"grammar_path" default:""`
GrammarString string `json:"grammar_string" default:""`
EnableJson bool `json:"enable_json" default:"false"`
EnableThink bool `json:"enable_think"`
TopK int32 `json:"top_k"`
MinP float32 `json:"min_p"`
RepetitionPenalty float32 `json:"repetition_penalty"`
GrammarPath string `json:"grammar_path"`
GrammarString string `json:"grammar_string"`
EnableJson bool `json:"enable_json"`

ChatCompletionNewParams
}

func defaultChatCompletionRequest() ChatCompletionRequest {
return ChatCompletionRequest{
Stream: false,
EnableThink: true,

TopK: 0,
MinP: 0.0,
RepetitionPenalty: 1.0,
GrammarPath: "",
GrammarString: "",
EnableJson: false,
}
}

var toolCallRegex = regexp.MustCompile(`<tool_call>([\s\S]+)<\/tool_call>` + "|" + "```json([\\s\\S]+)```")

// @Router /chat/completions [post]
Expand All @@ -53,7 +66,7 @@ var toolCallRegex = regexp.MustCompile(`<tool_call>([\s\S]+)<\/tool_call>` + "|"
// @Produce json
// @Success 200 {object} openai.ChatCompletion "Successful response for non-streaming requests."
func ChatCompletions(c *gin.Context) {
param := ChatCompletionRequest{}
param := defaultChatCompletionRequest()
if err := c.ShouldBindJSON(&param); err != nil {
c.JSON(http.StatusBadRequest, map[string]any{"error": err.Error()})
return
Expand Down Expand Up @@ -125,7 +138,7 @@ func chatCompletionsLLM(c *gin.Context, param ChatCompletionRequest) {
return
}
// Empty request for warm up
if len(param.Messages) == 0 {
if len(param.Messages) == 0 || (systemPrompt != "" && len(param.Messages) <= 1) {
c.JSON(http.StatusOK, nil)
return
}
Expand Down Expand Up @@ -355,7 +368,7 @@ func chatCompletionsVLM(c *gin.Context, param ChatCompletionRequest) {
}

// Empty request for warm up, just reset model state
if len(param.Messages) == 0 {
if len(param.Messages) == 0 || (systemPrompt != "" && len(param.Messages) <= 1) {
c.JSON(http.StatusOK, nil)
return
}
Expand Down Expand Up @@ -494,7 +507,7 @@ func parseSamplerConfig(param ChatCompletionRequest) *nexa_sdk.SamplerConfig {
TopP: float32(param.TopP.Value),
TopK: param.TopK,
MinP: param.MinP,
RepetitionPenalty: param.ReqetitionPenalty,
RepetitionPenalty: param.RepetitionPenalty,
PresencePenalty: float32(param.PresencePenalty.Value),
FrequencyPenalty: float32(param.FrequencyPenalty.Value),
Seed: int32(param.Seed.Value),
Expand Down