Update all references to canonical models (#29001)
* Script & Manual edition * Update
This commit is contained in:
@@ -89,8 +89,8 @@ class StreamerTester(unittest.TestCase):
|
||||
# Tests that we can pass `decode_kwargs` to the streamer to control how the tokens are decoded. Must be tested
|
||||
# with actual models -- the dummy models' tokenizers are not aligned with their models, and
|
||||
# `skip_special_tokens=True` has no effect on them
|
||||
tokenizer = AutoTokenizer.from_pretrained("distilgpt2")
|
||||
model = AutoModelForCausalLM.from_pretrained("distilgpt2").to(torch_device)
|
||||
tokenizer = AutoTokenizer.from_pretrained("distilbert/distilgpt2")
|
||||
model = AutoModelForCausalLM.from_pretrained("distilbert/distilgpt2").to(torch_device)
|
||||
model.config.eos_token_id = -1
|
||||
|
||||
input_ids = torch.ones((1, 5), device=torch_device).long() * model.config.bos_token_id
|
||||
|
||||
Reference in New Issue
Block a user