eeee

gpt.py

@@ -134,7 +134,7 @@ my_GPT = load_compiled_model_state_dict(my_GPT, 'latest_model_finetune.pth')
 #my_GPT.load_state_dict(torch.load('latest_model_finetune.pth', map_location=torch.device('cpu')))
 my_GPT.eval()
 
-my_GPT_code = GPT(enc.n_vocab, block_size, n_layers, n_heads, d_model, dropout=0.1) #enc.n_vocab
+my_GPT_code = GPT(enc.n_vocab, 256, n_layers, n_heads, d_model, dropout=0.1) #enc.n_vocab
 my_GPT_code = my_GPT_code.to(device)
 #my_GPT = torch.compile(my_GPT, mode='reduce-overhead')
 my_GPT_code = load_compiled_model_state_dict(my_GPT_code, 'mike-code-15k.pth')
@@ -153,8 +153,8 @@ def get_response(in_text, top_k=50, temperature=1, model="mike-chat"):
     prompt = "USER: " + in_text + "\nASSISTANT: "
     input_tokens = enc.encode(prompt)
     output_tokens = enc.encode(prompt)
-    for x in range(block_size):
-      if len(input_tokens) > block_size:
+    for x in range(models[model].block_size):
+      if len(input_tokens) > models[model].block_size:
         input_tokens = input_tokens[1:]
       context_tensor = torch.tensor(input_tokens).view(1, -1).to(device)
 

mike-code-15k.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3a3f92898af4acb8f41a571e6a4f2b597b7cca1316120787f51cb8d11ba84977
-size 650652710
+oid sha256:90941b8a7e50c178025368ba12120d700c1a4d7e84e39541d54c1e06a50d96ab
+size 649866278