Skip to content

Feat/llama cpp upgrade and support rocm image #1216

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
HaiHui886 merged 1 commit into from
Jun 26, 2026
Merged

Feat/llama cpp upgrade and support rocm image #1216

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
30 changes: 25 additions & 5 deletions builder/deploy/deployer.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -731,14 +731,17 @@ func (d *deployer) UpdateDeploy(ctx context.Context, dur *types.DeployUpdateReq,
deploy.EngineArgs = *dur.EngineArgs
}

if dur.Entrypoint != nil {
if deploy.RuntimeFramework == string(types.LlamaCpp) {
newVarStr, err := buildVariables(dur)
if isGGUFRuntime(deploy.RuntimeFramework) {
if dur.Entrypoint != nil {
newVarStr, err := buildVariables(deploy.Variables, dur)
if err != nil {
return fmt.Errorf("build variables for llama cpp error: %w", err)
}
dur.Variables = &newVarStr
}
if !hasGGUFEntryPoint(dur.Variables, deploy.Variables) {
return fmt.Errorf("entrypoint is required for llama.cpp or ktransformers")
}
}

if dur.Variables != nil {
Expand All @@ -761,8 +764,13 @@ func (d *deployer) UpdateDeploy(ctx context.Context, dur *types.DeployUpdateReq,
return nil
}

func buildVariables(dur *types.DeployUpdateReq) (string, error) {
varStr := ""
func isGGUFRuntime(runtimeFramework string) bool {
name := strings.ToLower(runtimeFramework)
return strings.Contains(name, string(types.LlamaCpp)) || strings.Contains(name, string(types.Ktransformers))
}

func buildVariables(existingVariables string, dur *types.DeployUpdateReq) (string, error) {
varStr := existingVariables
if dur.Variables != nil {
varStr = *dur.Variables
}
Expand All @@ -779,6 +787,18 @@ func buildVariables(dur *types.DeployUpdateReq) (string, error) {
return varStr, nil
}

func hasGGUFEntryPoint(variables *string, existingVariables string) bool {
varStr := existingVariables
if variables != nil {
varStr = *variables
}
varMap, err := hubcom.JsonStrToMap(varStr)
if err != nil {
return false
}
return strings.TrimSpace(varMap[types.GGUFEntryPoint]) != ""
}

func (d *deployer) StartDeploy(ctx context.Context, deploy *database.Deploy) error {
deploy.Status = common.Pending
// update deploy table
Expand Down
2 changes: 1 addition & 1 deletion component/model.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1216,7 +1216,7 @@ func (c *modelComponentImpl) Wakeup(ctx context.Context, namespace, name string,

func (c *modelComponentImpl) buildVariables(req types.ModelRunReq, frame *database.RuntimeFramework) (string, error) {
engineName := strings.ToLower(frame.FrameName)
if engineName == string(types.LlamaCpp) || engineName == string(types.Ktransformers) {
if strings.Contains(engineName, string(types.LlamaCpp)) || strings.Contains(engineName, string(types.Ktransformers)) {
//check entrypoint for llama.cpp
if len(req.Entrypoint) < 1 {
err := fmt.Errorf("entrypoint is required for llama.cpp or ktransformers")
Expand Down
157 changes: 157 additions & 0 deletions configs/inference/amd-llama.cpp.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,157 @@
{
"engine_name": "amd-llama.cpp",
"enabled": 1,
"model_format": "gguf",
"container_port": 8000,
"description": "llama.cpp OpenAI-compatible GGUF text generation service for ROCm",
"engine_images": [
{
"compute_type": "gpu",
"image": "opencsghq/llama.cpp-rocm:rocm7.2.2-b9787",
"driver_version": "7.2.2",
"engine_version": "b9787"
}
],
"engine_args": [
{ "name": "ctx-size", "value": "8192", "format": "-c %s" },
{ "name": "parallel", "value": "1", "format": "-np %s" },
{ "name": "batch-size", "value": "2048", "format": "-b %s" },
{ "name": "predict", "value": "-1", "format": "-n %s" },
{ "name": "threads", "value": "-1", "format": "-t %s" },
{ "name": "gpu-layers", "value": "-1", "format": "-ngl %s" },
{ "name": "custom-options", "value": "", "format": "%s" }
],
"supported_archs": [
"llama",
"llama4",
"deci",
"falcon",
"baichuan",
"grok",
"gpt2",
"gptj",
"gptneox",
"mpt",
"starcoder",
"refact",
"bert",
"modern-bert",
"nomic-bert",
"nomic-bert-moe",
"neo-bert",
"jina-bert-v2",
"jina-bert-v3",
"eurobert",
"bloom",
"stablelm",
"qwen",
"qwen2",
"qwen2moe",
"qwen2vl",
"qwen3",
"qwen3moe",
"qwen3next",
"qwen3vl",
"qwen3vlmoe",
"qwen35",
"qwen35moe",
"phi2",
"phi3",
"phimoe",
"plamo",
"plamo2",
"plamo3",
"codeshell",
"orion",
"internlm2",
"minicpm",
"minicpm3",
"gemma",
"gemma2",
"gemma3",
"gemma3n",
"gemma4",
"gemma4-assistant",
"gemma-embedding",
"starcoder2",
"rwkv6",
"rwkv6qwen2",
"rwkv7",
"arwkv7",
"mamba",
"mamba2",
"jamba",
"xverse",
"command-r",
"cohere2",
"cohere2moe",
"dbrx",
"olmo",
"olmo2",
"olmoe",
"openelm",
"arctic",
"deepseek",
"deepseek2",
"deepseek2-ocr",
"deepseek32",
"chatglm",
"glm4",
"glm4moe",
"glm-dsa",
"bitnet",
"t5",
"t5encoder",
"jais",
"jais2",
"nemotron",
"nemotron_h",
"nemotron_h_moe",
"exaone",
"exaone4",
"exaone-moe",
"granite",
"granitemoe",
"granitehybrid",
"chameleon",
"wavtokenizer-dec",
"plm",
"bailingmoe",
"bailingmoe2",
"dots1",
"arcee",
"afmoe",
"ernie4_5",
"ernie4_5-moe",
"falcon-h1",
"hunyuan-moe",
"hunyuan-dense",
"hunyuan_vl",
"smollm3",
"gpt-oss",
"lfm2",
"lfm2moe",
"dream",
"smallthinker",
"llada",
"llada-moe",
"seed_oss",
"grovemoe",
"apertus",
"minimax-m2",
"cogvlm",
"rnd1",
"pangu-embedded",
"mistral3",
"eagle3",
"mistral4",
"paddleocr",
"mimo2",
"step35",
"llama-embed",
"maincoder",
"kimi-linear",
"talkie",
"mellum"
]
}
90 changes: 74 additions & 16 deletions configs/inference/llama.cpp.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,20 +6,14 @@
"engine_images": [
{
"compute_type": "gpu",
"image": "opencsghq/llama.cpp:b5215",
"image": "opencsghq/llama.cpp:b9787",
"driver_version": "12.4",
"engine_version": "b5215"
},
{
"compute_type": "gpu",
"image": "opencsghq/llama.cpp:b5215-cu121",
"driver_version": "12.1",
"engine_version": "b5215"
"engine_version": "b9787"
},
{
"compute_type": "cpu",
"image": "opencsghq/llama.cpp:b5215-cpu",
"engine_version": "b5215"
"image": "opencsghq/llama.cpp:b9787-cpu",
"engine_version": "b9787"
}
],
"engine_args": [
Expand All @@ -36,18 +30,22 @@
"llama4",
"deci",
"falcon",
"baichuan",
"grok",
"gpt2",
"gptj",
"gptneox",
"mpt",
"baichuan",
"starcoder",
"refact",
"bert",
"modern-bert",
"nomic-bert",
"nomic-bert-moe",
"neo-bert",
"jina-bert-v2",
"jina-bert-v3",
"eurobert",
"bloom",
"stablelm",
"qwen",
Expand All @@ -56,10 +54,17 @@
"qwen2vl",
"qwen3",
"qwen3moe",
"qwen3next",
"qwen3vl",
"qwen3vlmoe",
"qwen35",
"qwen35moe",
"phi2",
"phi3",
"phimoe",
"plamo",
"plamo2",
"plamo3",
"codeshell",
"orion",
"internlm2",
Expand All @@ -68,11 +73,22 @@
"gemma",
"gemma2",
"gemma3",
"gemma3n",
"gemma4",
"gemma4-assistant",
"gemma-embedding",
"starcoder2",
"rwkv6",
"rwkv6qwen2",
"rwkv7",
"arwkv7",
"mamba",
"mamba2",
"jamba",
"xverse",
"command-r",
"cohere2",
"cohere2moe",
"dbrx",
"olmo",
"olmo2",
Expand All @@ -81,23 +97,65 @@
"arctic",
"deepseek",
"deepseek2",
"deepseek2-ocr",
"deepseek32",
"chatglm",
"glm4",
"glm4moe",
"glm-dsa",
"bitnet",
"t5",
"t5encoder",
"jais",
"jais2",
"nemotron",
"nemotron_h",
"nemotron_h_moe",
"exaone",
"rwkv6",
"rwkv6qwen2",
"rwkv7",
"arwkv7",
"exaone4",
"exaone-moe",
"granite",
"granitemoe",
"granitehybrid",
"chameleon",
"wavtokenizer-dec",
"plm",
"bailingmoe"
"bailingmoe",
"bailingmoe2",
"dots1",
"arcee",
"afmoe",
"ernie4_5",
"ernie4_5-moe",
"falcon-h1",
"hunyuan-moe",
"hunyuan-dense",
"hunyuan_vl",
"smollm3",
"gpt-oss",
"lfm2",
"lfm2moe",
"dream",
"smallthinker",
"llada",
"llada-moe",
"seed_oss",
"grovemoe",
"apertus",
"minimax-m2",
"cogvlm",
"rnd1",
"pangu-embedded",
"mistral3",
"eagle3",
"mistral4",
"paddleocr",
"mimo2",
"step35",
"llama-embed",
"maincoder",
"kimi-linear",
"talkie",
"mellum"
]
}
Loading
Loading