Use GemmaAttention for Gemma #72
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
qihqi
force-pushed
the
hanq_add_model
branch
2 times, most recently
from
ab0c882 to
5db75d6
Compare
This way it produces more accurate results
lsy323
approved these changes
May 9, 2024
convert_checkpoints.py
Outdated
|
|
||
| if new_key != key: | ||
| state_dict[new_key] = state_dict.pop(key) | ||
| output_ckpt_dir.mkdir(parents=True, exist_ok=True) |
Collaborator
There was a problem hiding this comment.
This shouldn't be needed. Output folder will be created in _export_to_local https://github.com/google/jetstream-pytorch/blob/811d718c1f93e5ce37182e2c1ec54d3dc0b4aed7/convert_checkpoints.py#L355
convert_checkpoints.py
Outdated
| ] | ||
| model_config = json.loads((input_ckpt_dir / "config.json").read_text()) | ||
| for key in list(state_dict.keys()): | ||
| print(key) |
FanhaiLu1
approved these changes
May 9, 2024
default_shardings/gemma.yaml
Outdated
|
|
||
| freqs_cis : -1 # torch.complex64 (16384, 128) | ||
| freqs_cis : null # torch.complex64 (16384, 128) | ||
| layers.*.self_attn.qkv_proj.weight: 0 |
Collaborator
There was a problem hiding this comment.
Is this one only for test purpose? In gemma model, I saw the code directly read wq,wk and wv.
| self.env.apply_sharding(output, axis=2) | ||
| return self.wo(output) | ||
| output = self.attention_kernel(xq, xk, xv, mask, cache) | ||
| output = output.transpose(-3, -2).contiguous().view(bsz, seqlen, -1) |
Collaborator
There was a problem hiding this comment.
Nice, code is cleaner with refactoring the attention kernel.
| return x_out | ||
|
|
||
|
|
||
| class GemmaAttention(nn.Module): |
Collaborator
There was a problem hiding this comment.
In long term, we might need to extend from Attention class, large percentage of code are similar.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This way it produces more accurate results (with EOS)
{'rouge1': 36.9881, 'rouge2': 13.3464, 'rougeL': 21.7437, 'rougeLsum': 35.1489, 'gen_len': 1295948, 'gen_num': 1000}