copy pasting what seems to work to bench,sample as well. ty @lantiga

bench.py

@@ -2,23 +2,29 @@
 A much shorter version of train.py for benchmarking
 """
 import os
+from contextlib import nullcontext
 import numpy as np
 import time
 import torch
 from model import GPTConfig, GPT
 
 # -----------------------------------------------------------------------------
-device = 'cuda'
 batch_size = 8
 block_size = 1024
-compile = True
+seed = 1337
+device = 'cuda' # examples: 'cpu', 'cuda', 'cuda:0', 'cuda:1', etc.
+dtype = 'bfloat16' # 'float32' or 'bfloat16' or 'float16'
+compile = True # use PyTorch 2.0 to compile the model to be faster
 exec(open('configurator.py').read()) # overrides from command line or config file
 # -----------------------------------------------------------------------------
 
-dtype = torch.bfloat16 # todo make configurable
+torch.manual_seed(seed)
+torch.cuda.manual_seed(seed)
 torch.backends.cuda.matmul.allow_tf32 = True # allow tf32 on matmul
 torch.backends.cudnn.allow_tf32 = True # allow tf32 on cudnn
-torch.manual_seed(1337)
+device_type = 'cuda' if 'cuda' in device else 'cpu' # for later use in torch.autocast
+ptdtype = {'float32': torch.float32, 'bfloat16': torch.bfloat16, 'float16': torch.float16}[dtype]
+ctx = nullcontext() if device_type == 'cpu' else torch.amp.autocast(device_type=device_type, dtype=ptdtype)
 
 # data loading init
 real_data = True
@@ -74,7 +80,7 @@ if profile:
 
         for k in range(num_steps):
             X, Y = get_batch('train')
-            with torch.autocast(device_type='cuda', dtype=dtype):
+            with ctx:
                 logits, loss = model(X, Y)
             optimizer.zero_grad(set_to_none=True)
             loss.backward()
@@ -92,7 +98,7 @@ else:
         t0 = time.time()
         for k in range(num_steps):
             X, Y = get_batch('train')
-            with torch.autocast(device_type='cuda', dtype=dtype):
+            with ctx:
                 logits, loss = model(X, Y)
             optimizer.zero_grad(set_to_none=True)
             loss.backward()

sample.py

@@ -2,20 +2,22 @@
 Sample from a trained model
 """
 import os
+from contextlib import nullcontext
 import torch
 import tiktoken
 from model import GPTConfig, GPT
 
 # -----------------------------------------------------------------------------
 out_dir = 'out'
-device = 'cuda'
-compile = False
 start = "\n" # or "<|endoftext|>" or whatever you like
 num_samples = 10 # number of samples to draw
 max_new_tokens = 500 # number of tokens generated in each sample
 temperature = 0.8 # higher temperature (up to 1) is more random, lower (down to 0) means more greedy
 top_k = 200 # retain only the top_k most likely tokens, clamp others to have 0 probability
 seed = 1337
+device = 'cuda' # examples: 'cpu', 'cuda', 'cuda:0', 'cuda:1', etc.
+dtype = 'bfloat16' # 'float32' or 'bfloat16' or 'float16'
+compile = False # use PyTorch 2.0 to compile the model to be faster
 exec(open('configurator.py').read()) # overrides from command line or config file
 # -----------------------------------------------------------------------------
 
@@ -23,6 +25,9 @@ torch.manual_seed(seed)
 torch.cuda.manual_seed(seed)
 torch.backends.cuda.matmul.allow_tf32 = True # allow tf32 on matmul
 torch.backends.cudnn.allow_tf32 = True # allow tf32 on cudnn
+device_type = 'cuda' if 'cuda' in device else 'cpu' # for later use in torch.autocast
+ptdtype = {'float32': torch.float32, 'bfloat16': torch.bfloat16, 'float16': torch.float16}[dtype]
+ctx = nullcontext() if device_type == 'cpu' else torch.amp.autocast(device_type=device_type, dtype=ptdtype)
 
 # model
 ckpt_path = os.path.join(out_dir, 'ckpt.pt')
@@ -45,11 +50,10 @@ enc = tiktoken.get_encoding("gpt2")
 start_ids = enc.encode(start, allowed_special={"<|endoftext|>"})
 x = (torch.tensor(start_ids, dtype=torch.long, device=device)[None, ...])
 
-for k in range(num_samples):
-
+# run generation
 with torch.no_grad():
-        with torch.amp.autocast(device_type="cuda", dtype=torch.bfloat16):
+    with ctx:
+        for k in range(num_samples):
             y = model.generate(x, max_new_tokens, temperature=temperature, top_k=top_k)
-
         print(enc.decode(y[0].tolist()))
         print('---------------')