Upload README.md with huggingface_hub

README.md

@@ -169,13 +169,14 @@ The model can generate TinyStories-like text fragments, but repetitions and temp
 
 ```python
 import torch
-from transformers import AutoTokenizer, Gemma4ForCausalLM
+from transformers import PreTrainedTokenizerFast, Gemma4ForCausalLM
 
 repo = "shibatch/tinygemma4text3m"
 
-tokenizer = AutoTokenizer.from_pretrained(repo)
+tokenizer = PreTrainedTokenizerFast.from_pretrained(repo, subfolder="hf")
 model = Gemma4ForCausalLM.from_pretrained(
     repo,
+    subfolder="hf",
     torch_dtype=torch.float32,
 )
 model.eval()
@@ -195,16 +196,6 @@ with torch.no_grad():
 print(tokenizer.decode(output_ids[0], skip_special_tokens=True))
 ```
 
-If the model files are stored under an `hf/` subdirectory in the repository, use:
-
-```python
-tokenizer = AutoTokenizer.from_pretrained(repo, subfolder="hf")
-model = Gemma4ForCausalLM.from_pretrained(
-    repo,
-    subfolder="hf",
-    torch_dtype=torch.float32,
-)
-```
 
 ## Loading with Transformers