Add a token counting endpoint for vLLM

This commit is contained in:
sasha0552
2024-06-20 01:13:32 +00:00
committed by GitHub
parent 584d0e6222
commit 6c87dfe10d
2 changed files with 8 additions and 4 deletions

View File

@ -7,7 +7,7 @@ import { kai_flags } from './kai-settings.js';
import { textgen_types, textgenerationwebui_settings as textgen_settings, getTextGenServer, getTextGenModel } from './textgen-settings.js';
import { getCurrentDreamGenModelTokenizer, getCurrentOpenRouterModelTokenizer, openRouterModels } from './textgen-models.js';
const { OOBA, TABBY, KOBOLDCPP, APHRODITE, LLAMACPP, OPENROUTER, DREAMGEN } = textgen_types;
const { OOBA, TABBY, KOBOLDCPP, VLLM, APHRODITE, LLAMACPP, OPENROUTER, DREAMGEN } = textgen_types;
export const CHARACTERS_PER_TOKEN_RATIO = 3.35;
const TOKENIZER_WARNING_KEY = 'tokenizationWarningShown';
@ -39,7 +39,7 @@ export const SENTENCEPIECE_TOKENIZERS = [
//tokenizers.NERD2,
];
export const TEXTGEN_TOKENIZERS = [OOBA, TABBY, KOBOLDCPP, LLAMACPP, APHRODITE];
export const TEXTGEN_TOKENIZERS = [OOBA, TABBY, KOBOLDCPP, LLAMACPP, VLLM, APHRODITE];
const TOKENIZER_URLS = {
[tokenizers.GPT2]: {
@ -765,6 +765,7 @@ function getTextgenAPITokenizationParams(str) {
api_type: textgen_settings.type,
url: getTextGenServer(),
legacy_api: textgen_settings.legacy_api && (textgen_settings.type === OOBA || textgen_settings.type === APHRODITE),
vllm_model: textgen_settings.vllm_model,
};
}