Update app.py

app.py

@@ -1,66 +1,66 @@
-import tiktoken
-import torch
-import biggerbrain as biggerbrain
-import ai_extras as A_E
-from safetensors.torch import save_file, load_file
-
-model = biggerbrain.initmodel("cpu")
-model = model._orig_mod if hasattr(model, '_orig_mod') else model
-
-while True:
-    user_input = input("You: ")
-    user_input = user_input.lower()
-    if user_input.lower() in {"exit", "quit"}:
-        print("Exiting the app.")
-        break
-    
-    elif user_input == "print model":
-        model.print_parameter_breakdown(model)
-    elif user_input == "cpu":
-        model.to("cpu")
-        print("Model moved to CPU.")
-    elif user_input == "train":
-        if 'pretrain_ds' not in locals():
-            pretrain_ds = A_E.StreamDataset(bin_file="C:\\AIs\\biggerbrain2_135m\\total_dataset.bin", seq_len=model.sequencelength)
-            print(f"Dataset loaded: {len(pretrain_ds)} samples") 
-        model.trainingloop(data=pretrain_ds, epochs=10, lr=3e-4, batchsize=4, accumulation_steps=32, warmup_steps=5000)#train
-    elif user_input.lower() == "load":
-        weights = load_file("best_model.safetensors")
-        model.load_state_dict(weights, strict=False)
-        print("Weights loaded!")
-    elif user_input.lower() == "check1":
-        print("alpha pre:")
-        print(model.get_parameter("alpha_pre").item())
-        print("alpha loop:")
-        print(model.get_parameter("alpha_loop").item())
-        print("alpha post:")
-        print(model.get_parameter("alpha_post").item())
-        print("alpha mem:")
-        print(model.get_parameter("alpha_mem").item())
-        model.debugprints = True
-        model.forward_training(1)
-    elif user_input.lower() == "check2":
-        print("\n--- Model Internal Stats ---")
-        # Print all alpha parameters dynamically
-        for name, param in model.named_parameters():
-            if 'alpha' in name:
-                # Using .item() to get the actual number instead of the tensor object
-                print(f"{name}: {param.item():.6f}")
-        
-        # Check the Engram Gate (mem_gate)
-        if hasattr(model, 'mem_gate'):
-            # We look at the bias because that's what controls the initial "openness"
-            gate_bias = model.mem_gate.bias.item()
-            # Calculate the actual sigmoid value to see the % it's open
-            gate_open_pct = torch.sigmoid(torch.tensor(gate_bias)).item() * 100
-            print(f"mem_gate bias: {gate_bias:.6f} ({gate_open_pct:.2f}% open)")
-        
-        print("----------------------------\n")
-    elif user_input.lower() == "check3":
-        for name, param in model.named_parameters():
-            if 'alpha' in name:
-                print(f"{name} | Requires Grad: {param.requires_grad} | Device: {param.device}")
-    elif user_input == "debug":
-        model.debugprints = True
-    else:
+import tiktoken
+import torch
+import biggerbrain as biggerbrain
+import ai_extras as A_E
+from safetensors.torch import save_file, load_file
+
+model = biggerbrain.initmodel("cpu")
+model = model._orig_mod if hasattr(model, '_orig_mod') else model
+
+while True:
+    user_input = input("You: ")
+    user_input = user_input.lower()
+    if user_input.lower() in {"exit", "quit"}:
+        print("Exiting the app.")
+        break
+    
+    elif user_input == "print model":
+        model.print_parameter_breakdown(model)
+    elif user_input == "cpu":
+        model.to("cpu")
+        print("Model moved to CPU.")
+    elif user_input == "train":
+        if 'pretrain_ds' not in locals():
+            pretrain_ds = A_E.StreamDataset(bin_file="C:\\AIs\\biggerbrain2_135m\\total_dataset.bin", seq_len=model.sequencelength)
+            print(f"Dataset loaded: {len(pretrain_ds)} samples") 
+        model.trainingloop(data=pretrain_ds, epochs=10, lr=3e-4, batchsize=4, accumulation_steps=32, warmup_steps=5000)#train
+    elif user_input.lower() == "load":
+        weights = load_file("model.safetensors")
+        model.load_state_dict(weights, strict=False)
+        print("Weights loaded!")
+    elif user_input.lower() == "check1":
+        print("alpha pre:")
+        print(model.get_parameter("alpha_pre").item())
+        print("alpha loop:")
+        print(model.get_parameter("alpha_loop").item())
+        print("alpha post:")
+        print(model.get_parameter("alpha_post").item())
+        print("alpha mem:")
+        print(model.get_parameter("alpha_mem").item())
+        model.debugprints = True
+        model.forward_training(1)
+    elif user_input.lower() == "check2":
+        print("\n--- Model Internal Stats ---")
+        # Print all alpha parameters dynamically
+        for name, param in model.named_parameters():
+            if 'alpha' in name:
+                # Using .item() to get the actual number instead of the tensor object
+                print(f"{name}: {param.item():.6f}")
+        
+        # Check the Engram Gate (mem_gate)
+        if hasattr(model, 'mem_gate'):
+            # We look at the bias because that's what controls the initial "openness"
+            gate_bias = model.mem_gate.bias.item()
+            # Calculate the actual sigmoid value to see the % it's open
+            gate_open_pct = torch.sigmoid(torch.tensor(gate_bias)).item() * 100
+            print(f"mem_gate bias: {gate_bias:.6f} ({gate_open_pct:.2f}% open)")
+        
+        print("----------------------------\n")
+    elif user_input.lower() == "check3":
+        for name, param in model.named_parameters():
+            if 'alpha' in name:
+                print(f"{name} | Requires Grad: {param.requires_grad} | Device: {param.device}")
+    elif user_input == "debug":
+        model.debugprints = True
+    else:
         biggerbrain.think(prompt=user_input, model=model, max_length=10, iter=3, top_k=10, temperature=1.0)