Update README.md

README.md

@@ -22,6 +22,13 @@ It is **not** a full base model. Please load it on top of the base model below.
 - `adapter_config.json` : LoRA config (PEFT)
 - `tokenizer.json`, `tokenizer_config.json`, `vocab.json`, `merges.txt` : tokenizer files
 - `chat_template.jinja` : chat template (if used)
+- > Note: This repository contains **LoRA adapter weights only**. You must download the base model (`Qwen/Qwen3-4B-Instruct-2507`) separately.
+
+
+## Requirements
+- `transformers` (Qwen3対応の版)
+- `peft`
+- `torch`
 
 ## How to load (Transformers + PEFT)
 
@@ -247,16 +254,18 @@ Carbon emissions can be estimated using the [Machine Learning Impact calculator]
 
 [More Information Needed]
 
+## Quick test generation
+```python
 ## Quick test generation
 ```python
 import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM
 from peft import PeftModel
 
-BASE_MODEL = "<Qwen/Qwen3-4B-Instruct-2507>"
+BASE_MODEL = "Qwen/Qwen3-4B-Instruct-2507"
 ADAPTER_REPO = "leaf0788/structeval-lora"
 
-tokenizer = AutoTokenizer.from_pretrained(ADAPTER_REPO, trust_remote_code=True)
+tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL, trust_remote_code=True)
 base = AutoModelForCausalLM.from_pretrained(
     BASE_MODEL,
     torch_dtype=torch.float16,
@@ -265,10 +274,12 @@ base = AutoModelForCausalLM.from_pretrained(
 )
 model = PeftModel.from_pretrained(base, ADAPTER_REPO).eval()
 
-prompt = "Please output JSON code.\n\nTask: Return a JSON with a single key `hello` and value `world`."
+prompt = 'Please output JSON code.\n\nTask: Return a JSON with a single key "hello" and value "world".'
 inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
 
 with torch.no_grad():
     out = model.generate(**inputs, max_new_tokens=128, do_sample=False)
 
-print(tokenizer.decode(out[0], skip_special_tokens=True))
+gen = tokenizer.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True)
+print(gen)
+