Update README.md

README.md

@@ -57,7 +57,7 @@ model.load_weights(th_p_, strict=True)
 ### Example: Evaluate the model on some text
 
 ```python
-def eval(text_: str, model, config, per_token=False):
+def eval_loss(text_: str, model, config, per_token=False):
     text_ = text_.encode('utf-8')
     
     x_prev_0s, state_prevs, x_prev_1s = (mx.zeros([config['layers'], 1, 1, config['input_dims']], dtype=dtype), 
@@ -82,7 +82,12 @@ def eval(text_: str, model, config, per_token=False):
         return nn.losses.cross_entropy(logits, mx.roll(txt_btch, -1, axis=1))[:, :-1].mean(), (mx.argmax(logits, axis=-1) == mx.roll(txt_btch, -1, axis=1)).mean()
 ```
 
-```python
+The text should show something like '[STX]def to_char(x): ...' since '[STX]' is my start token. Else, add the \x02 character in, NOT the picture version.
+
+The STX character should appear **bright red**, the version on the right is the correct one.
+![image/png](https://cdn-uploads.huggingface.co/production/uploads/66a767dcbe4c3c2683495a8b/rTVMKiioh-uo1Syim3BZ3.png)
+
+```python 
 text_ = '''def to_char(x):
     try:
         return bytes([x]).decode('utf-8')
@@ -90,11 +95,11 @@ text_ = '''def to_char(x):
         return f'{x}'
 '''
 
-print(eval(text_, model, config))
+print(eval_loss(text_, model, config)) # returns (CE Loss, Accuracy of next character)
 ```
 
 ```
-(array(0.738281, dtype=bfloat16), array(0.77451, dtype=float32)) # (CE Loss, Accuracy of next character)
+(array(0.738281, dtype=bfloat16), array(0.77451, dtype=float32))
 ```
 
 ### Example: Visualize the attention maps (beta)