Update PT_NVFP4Quant.md and PT_MXQuant.md

[xin3he]

User description

Improve the readability of the README files

---

PR Type

Enhancement, Documentation

---

Description

• Improve dequantization formula formatting

• Update quantization configuration comments

• Enhance readability of PT_NVFP4Quant.md

---

Diagram Walkthrough

flowchart LR
  A["Update dequantization formula"] -- "Use inline code" --> B["Enhance readability"]
  B -- "Add comments" --> C["Clarify quantization config"]

Loading

File Walkthrough

	Relevant files
Documentation	PT_NVFP4Quant.md Update dequantization formula and configuration comments  docs/source/3x/PT_NVFP4Quant.md Updated dequantization formula to use inline code Added comments to clarify quantization configuration parameters Improved overall readability of the document +7/-8

---

[PRAgent4INC]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 2 🔵🔵⚪⚪⚪

🧪 No relevant tests

🔒 No security concerns identified

⚡ Recommended focus areas for review Formula Formatting The change from LaTeX-style formula to inline code may reduce readability for users familiar with LaTeX. Consider using a consistent format throughout the document. `dequantized_value = quantized_value × block_scale × global_scale` Comment Clarity The comment for iters could be more precise. It mentions RTN without defining it, which might confuse readers unfamiliar with the term. # `iters=0` means RTN (fast, no optimization); use default `iters=200` if accuracy is poor quant_config = AutoRoundConfig( tokenizer=tokenizer, # Tokenizer for processing calibration data scheme="NVFP4", # NVFP4 quantization scheme iters=0, # Number of optimization iterations (default: 200)

[PRAgent4INC]

PR Code Suggestions ✨

Explore these optional code suggestions:

Category	Suggestion	Impact
General	Use LaTeX for math Use standard mathematical notation for better readability. docs/source/3x/PT_NVFP4Quant.md [51] -`dequantized_value = quantized_value × block_scale × global_scale` +$$\text{dequantized\_value} = \text{quantized\_value} \times \text{block\_scale} \times \text{global\_scale}$$ Suggestion importance[1-10]: 7 __ Why: Using LaTeX for mathematical notation improves readability, but since this is a markdown file, it might not render correctly in all viewers.	Medium
	Clarify default iterations Remove redundant comment and clarify default value. docs/source/3x/PT_NVFP4Quant.md [73-78] -# `iters=0` means RTN (fast, no optimization); use default `iters=200` if accuracy is poor quant_config = AutoRoundConfig( tokenizer=tokenizer, # Tokenizer for processing calibration data scheme="NVFP4", # NVFP4 quantization scheme - iters=0, # Number of optimization iterations (default: 200) + iters=0, # RTN (fast, no optimization); use 200 for default accuracy export_format="llm_compressor", # Export format for the quantized model output_dir=output_dir, # Directory to save the quantized model (default: "temp_auto_round") ) Suggestion importance[1-10]: 6 __ Why: The suggestion clarifies the default value of iters but keeps the comment, which is still useful for understanding the code.	Low