Fix Qwen and Command tokenizers not used for logit bias

This commit is contained in:
Cohee 2024-09-17 13:01:19 +00:00
parent 0b0bd27321
commit 0207794a2b
2 changed files with 7 additions and 3 deletions

View File

@ -15,7 +15,7 @@ import { BIAS_CACHE, createNewLogitBiasEntry, displayLogitBias, getLogitBiasList
import { power_user, registerDebugFunction } from './power-user.js';
import { getEventSourceStream } from './sse-stream.js';
import { getCurrentDreamGenModelTokenizer, getCurrentOpenRouterModelTokenizer } from './textgen-models.js';
import { SENTENCEPIECE_TOKENIZERS, TEXTGEN_TOKENIZERS, getTextTokens, tokenizers } from './tokenizers.js';
import { ENCODE_TOKENIZERS, TEXTGEN_TOKENIZERS, getTextTokens, tokenizers } from './tokenizers.js';
import { getSortableDelay, onlyUnique } from './utils.js';
export {
@ -353,7 +353,7 @@ function getTokenizerForTokenIds() {
return tokenizers.API_CURRENT;
}
if (SENTENCEPIECE_TOKENIZERS.includes(power_user.tokenizer)) {
if (ENCODE_TOKENIZERS.includes(power_user.tokenizer)) {
return power_user.tokenizer;
}

View File

@ -33,18 +33,22 @@ export const tokenizers = {
BEST_MATCH: 99,
};
export const SENTENCEPIECE_TOKENIZERS = [
// A list of local tokenizers that support encoding and decoding token ids.
export const ENCODE_TOKENIZERS = [
tokenizers.LLAMA,
tokenizers.MISTRAL,
tokenizers.YI,
tokenizers.LLAMA3,
tokenizers.GEMMA,
tokenizers.JAMBA,
tokenizers.QWEN2,
tokenizers.COMMAND_R,
// uncomment when NovelAI releases Kayra and Clio weights, lol
//tokenizers.NERD,
//tokenizers.NERD2,
];
// A list of Text Completion sources that support remote tokenization.
export const TEXTGEN_TOKENIZERS = [OOBA, TABBY, KOBOLDCPP, LLAMACPP, VLLM, APHRODITE];
const TOKENIZER_URLS = {