feat(readme): update README.md

Author: dineshpinto

README.md

@@ -12,14 +12,14 @@ Flare AI Kit template for Social AI Agents.
    | --------------------- | -------------------------------------------------------------------------- | ------------------------------------ |
    | `tuned_model_name`    | Name of the newly tuned model.                                             | `pugo-hilion`                        |
    | `tuning_source_model` | Name of the foundational model to tune on.                                 | `models/gemini-1.5-flash-001-tuning` |
-   | `epoch_count`         | Number of tuning epochs to run. An epoch is a pass over the whole dataset. | `100`                                |
+   | `epoch_count`         | Number of tuning epochs to run. An epoch is a pass over the whole dataset. | `30`                                 |
    | `batch_size`          | Number of examples to use in each training batch.                          | `4`                                  |
    | `learning_rate`       | Step size multiplier for the gradient updates.                             | `0.001`                              |
 
 ### Fine tuning a model over a dataset
 
 1. **Prepare a dataset:**
-   An example dataset is provided in `src/data/training_data.json`, which consists of tweet from
+   An example dataset is provided in `src/data/training_data.json`, which consists of tweets from
    [Hugo Philion's X](https://x.com/HugoPhilion) account. You can use any publicly available dataset
    for model fine-tuning.
 
@@ -33,8 +33,10 @@ Flare AI Kit template for Social AI Agents.
 3. **Observe loss parameters:**
    After tuning in complete, a training loss PNG will be saved in the root folder corresponding to the new model.
    Ideally the loss should minimize to near 0 after several training epochs.
+   
+   ![pugo-hilion_mean_loss](https://github.com/user-attachments/assets/f6c4d82b-678a-4ae5-bfb7-39dc59e1103d)
 
-4. **Test the new model**
+5. **Test the new model**
    Select the new tuned model and compare it against a set of prompting techniques (zero-shot, few-shot and chain-of-thought):
 
    ```bash