Upload folder using huggingface_hub

README.md

@@ -35,9 +35,12 @@ The architecture is a single-layer associative network trained via Hebbian learn
 ## Quick Start
 
 ```python
-from transformers import AutoModelForCausalLM
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+tokenizer = AutoTokenizer.from_pretrained("david-ar/20q", trust_remote_code=True)
 model = AutoModelForCausalLM.from_pretrained("david-ar/20q", trust_remote_code=True)
-model.play()
+model.set_vocab(tokenizer.questions, tokenizer.targets)
+model.play()  # interactive CLI game
 ```
 
 ## Pipeline Usage
@@ -49,6 +52,7 @@ from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
 
 tokenizer = AutoTokenizer.from_pretrained("david-ar/20q", trust_remote_code=True)
 model = AutoModelForCausalLM.from_pretrained("david-ar/20q", trust_remote_code=True)
+model.set_vocab(tokenizer.questions, tokenizer.targets)
 pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
 
 messages = [
@@ -82,7 +86,7 @@ The model is a weight matrix mapping 156 features (questions) to 1,200 output cl
 
 ## Why This Exists
 
-Mostly to see if it could be done. A 252KB model that plays a conversational guessing game, loaded through `from_pretrained`, running through `pipeline("text-generation")` with chat templates. Every bit of it works the same as models a million times its size.
+Mostly to see if it could be done. A 214KB model that plays a conversational guessing game, loaded through `from_pretrained`, running through `pipeline("text-generation")` with chat templates. Every bit of it works the same as models a million times its size.
 
 Also: 2-bit quantization was cool before it was cool.