README: decode with skip_special_tokens=True (EOS prefix is a doc boundary, not user text)
Browse files
README.md
CHANGED
|
@@ -41,7 +41,7 @@ tokenizer = AutoTokenizer.from_pretrained(
|
|
| 41 |
prompt = "The meaning of life is"
|
| 42 |
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
|
| 43 |
out = model.generate(**inputs, effort="high", max_new_tokens=64)
|
| 44 |
-
print(tokenizer.decode(out[0]))
|
| 45 |
```
|
| 46 |
|
| 47 |
With `pipeline`:
|
|
|
|
| 41 |
prompt = "The meaning of life is"
|
| 42 |
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
|
| 43 |
out = model.generate(**inputs, effort="high", max_new_tokens=64)
|
| 44 |
+
print(tokenizer.decode(out[0], skip_special_tokens=True))
|
| 45 |
```
|
| 46 |
|
| 47 |
With `pipeline`:
|