Update README.md

README.md

@@ -27,13 +27,13 @@ pipeline_tag: text-generation
 # NuMarkdown-reasoning 📄
 
 **NuMarkdown-reasoning** is the first reasoning vision-language model trained specifically to convert documents into clean GitHub-flavoured Markdown.
-It is a fine-tune of **Qwen 2.5-VL-7B** using ~10 k synthetic doc-to-Reasoning-to-Markdown pairs, followed by a RL phase (GRPO) with a layout-centric reward.
+It is a fine-tune of **Qwen 2.5-VL-7B** using ~10k synthetic Doc-to-Reasoning-to-Markdown pairs, followed by an RL phase (GRPO) with a layout-centric reward.
 
-*(note: the number of thinking tokens can vary from 20% to 2X the number of token of the final answers)*
+*(Note: the number of thinking tokens can vary from 20% to 2X the number of tokens of the final answers)*
 
 ## Results
 
-**NuMarkdown-reasoning** is significantly better than similar size non-reasoning models trained for markdown generation on complex documents, and achieve competitive results against top close sources alternatives.
+**NuMarkdown-reasoning** is significantly better than similar size non-reasoning models trained for markdown generation on complex documents, and achieves competitive results against top closed source alternatives.
 
 ### Arena ranking (using trueskill-2 ranking system):
 <p align="center">
@@ -50,7 +50,7 @@ It is a fine-tune of **Qwen 2.5-VL-7B** using ~10 k synthetic doc-to-Reasoning-t
 
 </p>
 
-*we plan to realease a markdown arena, similar to llmArena, for complex document to markdown task to help evaluate different document to markdown solution*
+*We plan to realease a markdown arena, similar to llmArena, for complex document-to-markdown tasks to help evaluate different document to markdown solutions.*
 
 ### Win-rate against others models (image-only):
 <p align="center">
@@ -60,10 +60,10 @@ It is a fine-tune of **Qwen 2.5-VL-7B** using ~10 k synthetic doc-to-Reasoning-t
 
 ## Training
 
-1. **SFT**: One-epoch supervised fine-tune on synthetic reasoning trace generated from public PDFs (10K input/output pairs).  
-2. **RL (GRPO)**: RL phase using a layout-centric reward (5K difficults image examples).
+1. **SFT**: One-epoch supervised fine-tuning on synthetic reasoning traces generated from public PDFs (10K input/output pairs).  
+2. **RL (GRPO)**: RL phase using a layout-centric reward (5K difficult image examples).
 
-**Model before GRPO loose 80% time vs post GRPO model (see win-rate matrix)**
+**Model before GRPO loses 80% time vs post-GRPO model (see win-rate matrix)**
 
 
 ## Quick start: 🤗 Transformers
@@ -107,7 +107,7 @@ print(processor.decode(out[0].split("<answer>")[1].split("</answer>")[0], skip_s
 ```
 
 
-## VLLM:
+## vLLM:
 ```python
 from PIL import Image
 from vllm import LLM, SamplingParams