Skip to content

Adds vLLM as Option for Local App #693

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 13 commits into from
Sep 4, 2024
Merged

Adds vLLM as Option for Local App #693

Show file tree
Hide file tree
Changes from 6 commits
Commits
Show all changes
13 commits
Select commit Hold shift + click to select a range
7dbd31d
Update local-apps.ts
EliMCosta May 21, 2024
32560d8
Update packages/tasks/src/local-apps.ts
EliMCosta May 22, 2024
f97747a
Update packages/tasks/src/local-apps.ts
EliMCosta May 22, 2024
82233de
Update local-apps.ts
EliMCosta May 22, 2024
2123430
Validation for `config.quantization_config.quant_method`
julien-c May 22, 2024
43de0e9
Update local-apps.ts
EliMCosta May 22, 2024
8557110
Update packages/tasks/src/local-apps.ts
EliMCosta May 30, 2024
2c3c6c2
Merge branch 'main' into patch-4
krampstudio Sep 2, 2024
17ad182
fix: udpate snippets
krampstudio Sep 3, 2024
2bb1bc1
Update packages/tasks/src/local-apps.ts
krampstudio Sep 4, 2024
d63b7cb
fix: rely only on gguf tag
krampstudio Sep 4, 2024
6fd56ef
Merge branch 'main' into patch-4
krampstudio Sep 4, 2024
0308303
Merge branch 'main' into patch-4
pcuenca Sep 4, 2024
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
64 changes: 62 additions & 2 deletions packages/tasks/src/local-apps.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,8 +42,29 @@ export type LocalApp = {
}
);

function isGgufModel(model: ModelData) {
return model.tags.includes("gguf");
function isGgufModel(model: ModelData): boolean {
return model.config?.quantization_config?.quant_method === "gguf";
}

function isAwqModel(model: ModelData): boolean {
return model.config?.quantization_config?.quant_method === "awq";
}

function isGptqModel(model: ModelData): boolean {
return model.config?.quantization_config?.quant_method === "gptq";
}

function isAqlmModel(model: ModelData): boolean {
return model.config?.quantization_config?.quant_method === "aqlm";
}

function isMarlinModel(model: ModelData): boolean {
return model.config?.quantization_config?.quant_method === "marlin";
}

function isFullModel(model: ModelData): boolean {
// Assuming a full model is identified by not having a quant_method
return !model.config?.quantization_config?.quant_method;
Copy link
Collaborator

@krampstudio krampstudio Jun 4, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

isFullModel creates a lot of false positive.

Instead we can maybe check against the supported architectures as suggested by @simon-mo

Something like this

const VLLM_SUPPORTED_ARCHS = [
    "AquilaForCausalLM", "ArcticForCausalLM", "BaiChuanForCausalLM", "BloomForCausalLM", ...
];
model.config?.architectures?.some((arch) => VLLM_SUPPORTED_ARCHS.includes(arch)

cc @julien-c

Copy link
Contributor

@mgoin mgoin Jun 13, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You can query the vLLM package for this list once you have it installed:

>>> from vllm import ModelRegistry
>>> ModelRegistry.get_supported_archs()
['AquilaModel', 'AquilaForCausalLM', 'BaiChuanForCausalLM', 'BaichuanForCausalLM', 'BloomForCausalLM', 'ChatGLMModel', 'ChatGLMForConditionalGeneration', 'CohereForCausalLM', 'DbrxForCausalLM', 'DeciLMForCausalLM', 'DeepseekForCausalLM', 'FalconForCausalLM', 'GemmaForCausalLM', 'GPT2LMHeadModel', 'GPTBigCodeForCausalLM', 'GPTJForCausalLM', 'GPTNeoXForCausalLM', 'InternLMForCausalLM', 'InternLM2ForCausalLM', 'JAISLMHeadModel', 'LlamaForCausalLM', 'LlavaForConditionalGeneration', 'LlavaNextForConditionalGeneration', 'LLaMAForCausalLM', 'MistralForCausalLM', 'MixtralForCausalLM', 'QuantMixtralForCausalLM', 'MptForCausalLM', 'MPTForCausalLM', 'MiniCPMForCausalLM', 'OlmoForCausalLM', 'OPTForCausalLM', 'OrionForCausalLM', 'PhiForCausalLM', 'Phi3ForCausalLM', 'QWenLMHeadModel', 'Qwen2ForCausalLM', 'Qwen2MoeForCausalLM', 'RWForCausalLM', 'StableLMEpochForCausalLM', 'StableLmForCausalLM', 'Starcoder2ForCausalLM', 'ArcticForCausalLM', 'XverseForCausalLM', 'Phi3SmallForCausalLM', 'MistralModel']

}

const snippetLlamacpp = (model: ModelData): string[] => {
Expand All @@ -63,6 +84,38 @@ LLAMA_CURL=1 make
];
};

const snippetVllm = (model: ModelData): string[] => {
return [
`
## Deploy with linux and docker (needs Docker installed) a gated model (please, request access in Hugginface's model repo):
docker run --runtime nvidia --gpus all \
--name my_vllm_container \
-v ~/.cache/huggingface:/root/.cache/huggingface \
--env "HUGGING_FACE_HUB_TOKEN=<secret>" \
-p 8000:8000 \
--ipc=host \
vllm/vllm-openai:latest \
--model ${model.id}
`,
`
## Load and run the model
docker exec -it my_vllm_container bash -c "python -m vllm.entrypoints.openai.api_server --model ${model.id} --dtype auto --api-key token-abc123"
`,
`
## Call the server using curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
-H "Content-Type: application/json" \
-H "Authorization: Bearer token-abc123" \
--data '{
"model": "'${model.id}'",
"messages": [
{"role": "user", "content": "Hello!"}
]
}'
`,
];
};

/**
* Add your new local app here.
*
Expand All @@ -82,6 +135,13 @@ export const LOCAL_APPS = {
displayOnModelPage: isGgufModel,
snippet: snippetLlamacpp,
},
vllm: {
prettyLabel: "vLLM",
docsUrl: "https://docs.vllm.ai",
mainTask: "text-generation",
displayOnModelPage: (model: ModelData) => isAwqModel(model) || isGptqModel(model) || isAqlmModel(model) || isMarlinModel(model) || isFullModel(model),
snippet: snippetVllm,
},
lmstudio: {
prettyLabel: "LM Studio",
docsUrl: "https://lmstudio.ai",
Expand Down
4 changes: 4 additions & 0 deletions packages/tasks/src/model-data.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,6 +52,10 @@ export interface ModelData {
bits?: number;
load_in_4bit?: boolean;
load_in_8bit?: boolean;
/**
* awq, gptq, aqlm, marlin, … Used by vLLM
*/
quant_method?: string;
};
tokenizer_config?: TokenizerConfig;
adapter_transformers?: {
Expand Down