islm / README.md

Upload folder using huggingface_hub

d2b17a2 verified 24 days ago

893 Bytes

base_model: Qwen/Qwen3-4B-Instruct-2507
library_name: peft
tags:
  - qlora
  - lora
  - comprehensible-input
  - language-learning
  - i-plus-1

i+1 Story SLM — QLoRA adapter

LoRA adapter over Qwen/Qwen3-4B-Instruct-2507 that writes comprehensible-input (i+1) language-learning stories in English, Chinese, and Japanese: every story stays inside a known-vocabulary set, adds at most one new word per sentence, recurs each target word >=3x, keeps new words inferable, and reads as a real story. See the dataset and eval harness for the behavior spec and base-vs-tuned numbers.

Load

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
base = "Qwen/Qwen3-4B-Instruct-2507"
tok = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(base, device_map="auto")
model = PeftModel.from_pretrained(model, "i0445/islm")