Update README.md

README.md

@@ -72,9 +72,9 @@ To set up the MelodyT5 environment and install the necessary dependencies, follo
 Before running the inference script, you can configure the following parameters in `config.py` or directly via command-line arguments:
 
 - `-num_tunes`: Number of independently computed returned tunes (default: 3)
-- `-max_patch`: Maximum length in tokens of each tune (default: 128)
-- `-top_p`: Tokens within the sample operation of text generation (default: 0.8)
-- `-top_k`: Tokens within the sample operation of text generation (default: 8)
+- `-max_patch`: Maximum length in patches of each tune (default: 128)
+- `-top_p`: Cumulative probability threshold for nucleus sampling (default: 0.8)
+- `-top_k`: Number of highest probability tokens to keep for sampling (default: 8)
 - `-temperature`: Temperature of the sampling operation (default: 2.6)
 - `-seed`: Seed for random state (default: None)
 - `-show_control_code`: Whether to show control codes (default: True)
@@ -86,7 +86,7 @@ These parameters control how the model generates melodies based on the input pro
 To perform inference tasks using MelodyT5, follow these steps:
 
 1. **Prepare Your Prompt**
-   - Edit `prompt.txt` to specify the task and input for the model. Each line in `prompt.txt` should contain a single prompt.
+   - Edit `prompt.txt` to specify the task and input for the model.
 
 2. **Execute Inference**
    - Run the following command to execute the inference script:
@@ -94,11 +94,11 @@ To perform inference tasks using MelodyT5, follow these steps:
      python inference.py -num_tunes 3 -max_patch 128 -top_p 0.8 -top_k 8 -temperature 2.6 -seed <seed_value> -show_control_code True
      ```
      Replace `<seed_value>` with your chosen seed value or leave it as `None` for a random seed.
-   - The script will generate melodies based on the prompts specified in `prompt.txt` using the configured parameters and save the results in the `output_tunes` folder.
+   - The script will generate melodies based on the prompt specified in `prompt.txt` using the configured parameters and save the results in the `output_tunes` folder.
 
 ## How to Use
 
-Follow these steps to effectively utilize MelodyT5 for symbolic music processing:
+Follow these steps to utilize MelodyT5 for symbolic music processing:
 
 1. **Prepare XML Data**
 
@@ -137,8 +137,6 @@ Follow these steps to effectively utilize MelodyT5 for symbolic music processing
    
    To perform inference tasks such as melody generation or harmonization, execute `inference.py`. The script reads prompts from `prompt.txt` to specify the task and input for the model. Customize prompts in `prompt.txt` to define different tasks and inputs for MelodyT5. Refer to the examples below for guidance on setting up prompts.
 
-   Ensure the input (encoder input) is complete, while the output (decoder input) is optional. If you need the model to continue a given output, use `%%input` and `%%output` to mark the beginning of each section. Additionally, the output must not contain incomplete bars. Here is an example prompt:
-
    ```
    %%input
    %%variation
@@ -155,6 +153,8 @@ Follow these steps to effectively utilize MelodyT5 for symbolic music processing
    |: B |
    ```
    
+   Ensure the input (encoder input) is complete, while the output (decoder input) is optional. If you need the model to continue a given output, use `%%input` and `%%output` to mark the beginning of each section. Additionally, the output must not contain incomplete bars.
+
 ## Inference Examples
 Below are the MelodyT5 results on seven MelodyHub tasks, using random samples from the validation set. Three independent outputs were generated without cherry-picking. Each `X:0` output corresponds to the original input for that task and is not generated by the model, while `X:1`, `X:2`, and `X:3` are generated outputs.