Update README.md

README.md

@@ -8,7 +8,7 @@ tags:
 - trl
 license: apache-2.0
 language:
-- en
+- ja
 ---
 
 # Uploaded  model
@@ -20,3 +20,109 @@ language:
 This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
 
 [<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
+
+---
+
+# How to use
+
+There are the normal steps from sample codes.
+
+0. ready to (you can skip this step in Google Colaboratry. )
+
+```shell
+# conda環境の構築
+wget "https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-$(uname)-$(uname -m).sh"  
+
+# このコマンドではいくつか質問があるので答えて下さい。おそらくインストール先のデフォルトは/root/miniforge3かと思います  
+bash Miniforge3-$(uname)-$(uname -m).sh  
+
+# 以下、インストール先が/root/miniforge3であることを前提とします  
+export PATH=/root/miniforge3/bin:$PATH  
+conda init  
+
+# ここで一度、terminalを立ち上げ直す必要があります。  
+# 以下のリンク先に従い環境を作ります。  
+# https://docs.unsloth.ai/get-started/installation/conda-install  
+conda create --name unsloth_env python=3.10 pytorch-cuda=12.1 pytorch cudatoolkit xformers -c pytorch -c nvidia -c xformers -y  
+conda activate unsloth_env  
+pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"  
+pip install --no-deps "trl<0.9.0" peft accelerate bitsandbytes  
+
+# jupyter notebook用のセットアップ。  
+conda install -c conda-forge ipykernel  
+python -m ipykernel install --user --name=unsloth_env --display-name "Python (unsloth_env)"  
+```
+
+## Follow these steps, run in the notebook:
+
+1. load model
+```shell
+%%capture
+!pip install unsloth
+!pip uninstall unsloth -y && pip install --upgrade --no-cache-dir "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"
+```
+
+```python
+from unsloth import FastLanguageModel
+import torch
+import json
+
+model_name = "tomofusa/llm-jp-3-13b-finetune-2"
+
+max_seq_length = 2048
+dtype = None
+load_in_4bit = True
+
+model, tokenizer = FastLanguageModel.from_pretrained(
+    model_name = model_name,
+    max_seq_length = max_seq_length,
+    dtype = dtype,
+    load_in_4bit = load_in_4bit,
+    # token = "hf-token", # In the Google Colab case, it call from ENV. If you want to write the token directly, please comment it out.
+)
+FastLanguageModel.for_inference(model)
+```
+
+3. Set up datasets and run inference.
+
+- Upload elyza-tasks-100-TV_0.jsonl to your workspace in manual.
+
+```python
+datasets = []
+with open("./elyza-tasks-100-TV_0.jsonl", "r") as f:
+    item = ""
+    for line in f:
+      line = line.strip()
+      item += line
+      if item.endswith("}"):
+        datasets.append(json.loads(item))
+        item = ""
+```
+
+```python
+from tqdm import tqdm
+
+# inference
+results = []
+for dt in tqdm(datasets):
+  input = dt["input"]
+
+  prompt = f"""### 指示\n{input}\n### 回答\n"""
+
+  inputs = tokenizer([prompt], return_tensors = "pt").to(model.device)
+
+  outputs = model.generate(**inputs, max_new_tokens = 512, use_cache = True, do_sample=False, repetition_penalty=1.2)
+  prediction = tokenizer.decode(outputs[0], skip_special_tokens=True).split('\n### 回答')[-1]
+
+  results.append({"task_id": dt["task_id"], "input": input, "output": prediction})
+```
+
+4. Save results to jsonl.
+
+```python
+file_name = model_name.replace("/", "_") + "_output.jsonl"
+with open(f"./{file_name}", 'w', encoding='utf-8') as f:
+    for result in results:
+        json.dump(result, f, ensure_ascii=False)
+        f.write('\n')
+```