cleanup brainfarts

app.py

@@ -104,26 +104,6 @@ def insert_fn(inpt: str, max_tokens, min_tokens, alpha_f, alpha_p, num_tokens_in
         print(e)
         yield ("Error...", gr.Text.update(value=str(e), visible=True))
 
-# def classify_fn_inner(inpt, clas):
-#     state = rwkv_rs.State(model)
-#     tokens = tokenizer.encode(f"This is an example of {clas} text: {inpt}").ids
-#     for i in tokens[:-2]:
-#         model.forward_token_preproc(i, state)
-#     # state_2 = state.copy()
-
-#     logit_x_1 = softmax(model.forward_token(tokens[-2], state))
-#     logit_y_1 = softmax(model.forward_token(tokens[-1], state))
-#     # shapep = logit_x_1.shape[0] * 0.9
-#     # s = np.sort(logit_y_1)[::-1]
-#     # c = s[np.argmax(np.cumsum(s) > 0.9)]
-#     # logit_y_1[logit_y_1 < c] = 0
-#     loss_1 = -np.sum(logit_y_1 * np.log(logit_x_1)) / logit_x_1.shape[0]
-
-#     # I forgor that I do not return the preproc shit...
-#     # logit_x_2 = model.forward_token_preproc(tokens[-2], state_2)
-#     # logit_y_2 = model.forward_token_preproc(tokens[-1], state_2)
-#     return (loss_1, None)
-
 def classify_fn_inner2(inpt, clas):
     state = rwkv_rs.State(model)
     tokens = tokenizer.encode(f"This is an example of {clas} text:").ids
@@ -135,7 +115,7 @@ def classify_fn_inner2(inpt, clas):
     for i in tokens[:-1]:
         logits.append(model.forward_token(i, state))
     logit_x = [softmax(i) for i in logits]
-    loss = -np.sum([ x[y] for x, y in zip(logit_x, tokens[1:]) ]) / len(logit_x)
+    loss = -np.sum([ np.log(x[y]) for x, y in zip(logit_x, tokens[1:]) ]) / len(logit_x)
 
     return loss 
 
@@ -143,23 +123,13 @@ def softmax(x):
     e = np.exp(x - np.max(x))
     return e / e.sum()
 
-# TODO: maybe make a function with pos/neg inputs?
 def classify_fn(inpt, clas, clasneg):
-    # loss_1, loss_2 = classify_fn_inner(inpt, clas)
-    # loss_1_neg, loss_2_neg = classify_fn_inner(inpt, clasneg)
-
-    # print(loss_1, loss_1_neg, end=' | ')
-    # # We negate the loss because we want to know who's closer to 0
-    # loss_1, loss_1_neg = softmax([-loss_1, -loss_1_neg])
-    # print(loss_1, loss_1_neg)
-
     loss_3 = classify_fn_inner2(inpt, clas)
     loss_3_neg = classify_fn_inner2(inpt, clasneg)
-    print(loss_3, loss_3_neg, end=' | ')
+    # print(loss_3, loss_3_neg, end=' | ')
     loss_3, loss_3_neg = softmax([-loss_3, -loss_3_neg])
-    print(loss_3, loss_3_neg)
+    # print(loss_3, loss_3_neg)
 
-    # return ({"v1_pos": loss_1, "v1_neg": loss_1_neg, "v3_pos": loss_3, "v3_neg": loss_3_neg})
     return ({"+": loss_3, "-": loss_3_neg})
 
 def generator_wrap(l, fn):