Update app.py

app.py

@@ -8,11 +8,40 @@ from functools import partial
 
 import gradio as gr
 import numpy as np
+import keras
+from keras import layers
 embedder = NVIDIAEmbeddings(model="nvolveqa_40k", model_type="query")
 chat_model = ChatNVIDIA(model="llama2_13b") | StrOutputParser()
 
 response_prompt = ChatPromptTemplate.from_messages([("system", "{system}"), ("user", "{input}")])
 
+
+
+def train_model_neural_network(class0, class1):
+    ## Classic deep learning training loop. If using this, train it to convergence
+    model = keras.Sequential([
+        layers.Dense(64, activation='tanh'),
+        layers.Dense(1, activation='sigmoid'),
+    ])
+    ## Since this network is so shallow and the embedding backbone is "kept frozen"
+    ##  a high learning rate should not overfit and will actually converge very quickly.
+    model.compile(
+        optimizer = keras.optimizers.Adam(learning_rate = 1),
+        loss = [keras.losses.BinaryCrossentropy(from_logits=False)],
+        metrics = [keras.metrics.BinaryAccuracy()],
+    )
+    ## Since this uses stochastic gradient descent, we'll need to repeat this process
+
+    reps_per_batch = 64*5  ## <- repeat the dataset, effectively increasing "epochs" without printing too much
+    epochs = 2             ## <- one epoch should actually be sufficient; 2 to print out an updated training loss
+    x = np.array((class0 + class1) * reps_per_batch)
+    y = np.array(([0]*len(class0) + [1]*len(class1)) * reps_per_batch)
+    model.fit(x, y, epochs=epochs, batch_size=64, validation_split=.5)
+    return model
+
+
+model1 = train_model_neural_network(poor_embeds, good_embeds)
+
 def RPrint(preface=""):
     def print_and_return(x, preface=""):
         print(f"{preface}{x}")