BLOOM minor fixes small test#18175
Conversation
- add correct revision - corrected dosctring for test - removed a test
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com> Co-authored-by: Nouamane Tazi <nouamane98@gmail.com>
|
The documentation is not available anymore as the PR was closed or merged. |
|
|
||
| path_350m = "bigscience/bloom-350m" | ||
| model = BloomForCausalLM.from_pretrained(path_350m, use_cache=True).cuda() | ||
| model = BloomForCausalLM.from_pretrained(path_350m, use_cache=True, revision="gs555750").cuda() |
There was a problem hiding this comment.
What does the revision argument do? Are we sure we want to add that?
There was a problem hiding this comment.
yes, this argument loads the model from the initial model we were using when designing those tests: https://huggingface.co/bigscience/bloom-350m/tree/gs555750
@Muennighoff recently wanted to push the newest weights of 350m so just adding the revision flag should prevent us breaking those tests
| tokenizer.decode(greedy_output_without_pad[0, :-3], skip_special_tokens=True), | ||
| ) | ||
|
|
||
| @slow |
There was a problem hiding this comment.
Could you just explain why this test is removed please?
There was a problem hiding this comment.
Following a discussion that we had on Slack, it appears that one should always use padding_side=left
Here is a link that quickly summarizes the issue: #17963 (comment)
| tokenizer.decode(greedy_output_without_pad[0, :-3], skip_special_tokens=True), | ||
| ) | ||
|
|
||
| @slow |
* minor fixes - add correct revision - corrected dosctring for test - removed a test * contrib credits Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com> Co-authored-by: Nouamane Tazi <nouamane98@gmail.com> Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com> Co-authored-by: Nouamane Tazi <nouamane98@gmail.com>
4 participants
Small modifications
cc @ydshieh @NouamaneTazi @Muennighoff