Update README.md

README.md

@@ -34,19 +34,28 @@ We've modified Flax's 'lm1b' example to train on Japanese dataset. You can find
 
 ## Usage
 
-Please refer to [this](https://cloud.google.com/tpu/docs/run-calculation-jax) document, create a TPU-VM instance, and then execute the following steps while performing shell access to that TPU-VM instance via SSH.
+Here, we explain the procedure to generate text from pretrained weights using a CPU. We used the following instance on GCE for the Python 3.8 environment.
 
-* [Run a calculation on a Cloud TPU VM using JAX](https://cloud.google.com/tpu/docs/run-calculation-jax)
+* Machine Type: c2-standard-4 (4 CPUs, 16GB Memory)
+* Disk: 100GB (Standard Persistent Disk)
+* OS: Ubuntu 20.04 LTS x86/64
+
+Install Python 3.8 and pip.
+
+```
+$ sudo apt-get update
+$ sudo apt-get install python3.8 python3-pip build-essential
+```
 
 ```
-(tpu-vm)$ pip install --upgrade huggingface_hub
+$ pip install --upgrade huggingface_hub
 ```
 
 Run the Python interpreter and download the model files.
 
 ```
-(tpu-vm)$ cd $HOME
-(tpu-vm)$ python3
+$ cd $HOME
+$ python3
 ```
 
 ```python
@@ -58,16 +67,18 @@ Run the Python interpreter and download the model files.
 Clone the source code and install the necessary Python packages.
 
 ```
-(tpu-vm)$ git clone -b 1.0.0.RC2 https://github.com/FookieMonster/transformer-lm-japanese
-(tpu-vm)$ cd ./transformer-lm-japanese/transformer_lm
-(tpu-vm)$ pip install -r requirements.txt
-(tpu-vm)$ pip install "jax[tpu]==0.3.2" -f https://storage.googleapis.com/jax-releases/libtpu_releases.html
+$ git clone -b 1.0.0.RC2 https://github.com/FookieMonster/transformer-lm-japanese
+$ cd ./transformer-lm-japanese/transformer_lm
+$ pip install -r requirements.txt
+$ pip install jax[cpu]==0.3.2
+$ pip install chex==0.1.5
+$ pip install protobuf==3.20.3
 ```
 
 Text generation using downloaded model files.
 
 ```
-(tpu-vm)$ python3 generate_text.py --workdir=$HOME/logs/japanese_0.1b_v1 \
+$ python3 generate_text.py --workdir=$HOME/logs/japanese_0.1b_v1 \
     --config=configs/japanese_0.1b_v1.py \
     --config.sampling_temperature=0.6 \
     --config.sampling_top_k=20 \