Update README.md

README.md

@@ -8,7 +8,7 @@ language:
 
 # TohokuNLP BERT-alpha 500M
 
-長系列 (4096, 8192 トークン) の入力を可能にした日本語 [BERT](https://aclanthology.org/N19-1423/) モデルです。
+長系列 (4,096, 8,192 トークン) の入力を可能にした日本語 [BERT](https://aclanthology.org/N19-1423/) モデルです。
 
 ## 利用方法
 
@@ -90,7 +90,7 @@ Whole Word Masking 単語分割器には、[vibrato](https://github.com/daac-too
 | Batch Size (tokens) | 1,146,880 | 2,293,760 |
 | Max Learning Rate | 1.0E-4 | 1.0E-4 |
 | Min Learning Rate | 1.0E-6 | N/A |
-| Learning Rate Warmup Steps | 10000 | N/A |
+| Learning Rate Warmup Steps | 10,000 | N/A |
 | Scheduler | cosine | constant |
 | Optimizer | AdamW | AdamW |
 | Optimizer Config | beta_1 = 0.9, beta_2 = 0.999, eps = 1.0E-8 | beta_1 = 0.9, beta_2 = 0.999, eps = 1.0E-8 |
@@ -155,7 +155,7 @@ Whole Word Masking 単語分割器には、[vibrato](https://github.com/daac-too
 
 # TohokuNLP BERT-alpha 500M
 
-A Japanese [BERT](https://aclanthology.org/N19-1423/) model capable of processing long sequences (4096, 8192 tokens).
+A Japanese [BERT](https://aclanthology.org/N19-1423/) model capable of processing long sequences (4,096, 8,192 tokens).
 
 ## Usage
 
@@ -234,7 +234,7 @@ We only implemented Masked Language Modeling (MLM) during training, without Next
 | Batch Size (tokens) | 1,146,880 | 2,293,760 |
 | Max Learning Rate | 1.0E-4 | 1.0E-4 |
 | Min Learning Rate | 1.0E-6 | N/A |
-| Learning Rate Warmup Steps | 10000 | N/A |
+| Learning Rate Warmup Steps | 10,000 | N/A |
 | Scheduler | cosine | constant |
 | Optimizer | AdamW | AdamW |
 | Optimizer Config | beta_1 = 0.9, beta_2 = 0.999, eps = 1.0E-8 | beta_1 = 0.9, beta_2 = 0.999, eps = 1.0E-8 |