Conversation
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
ArthurZucker
left a comment
ArthurZucker
left a comment
There was a problem hiding this comment.
Super nice, you are missing the test files, integration tests etc! (And readme etc)
| initializer_range=0.02, | ||
| rms_norm_eps=0.00000015625, | ||
| use_rms_norm=True, | ||
| apply_residual_connection_post_layernorm=False, |
There was a problem hiding this comment.
is this false for all models? If so, to delete!
| self.mlp = GlmMLP(config) | ||
| self.input_layernorm = ( | ||
| GlmRMSNorm(config.hidden_size, eps=config.rms_norm_eps) | ||
| if config.use_rms_norm |
There was a problem hiding this comment.
check what config uses, but we avoid that in general as well! (code path)
| """ | ||
|
|
||
| hidden_states_after_norm = self.input_layernorm(hidden_states) | ||
| residual = hidden_states_after_norm if self.apply_residual_connection_post_layernorm else hidden_states |
There was a problem hiding this comment.
same here! check if any released models have both
| self.layers = nn.ModuleList( | ||
| [GlmDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)] | ||
| ) | ||
| if config.post_layer_norm: |
There was a problem hiding this comment.
Init should be like this: https://github.com/huggingface/transformers/pull/31329/files#diff-e13de4b5db0c6872b5f0ec197d07fdaf80174b37f55bd4fefbe1526e57635683
(you could not know but we ought to enforce this now!)
* HQQ model serialization attempt * fix hqq dispatch and unexpected keys * style * remove check_old_param * revert to check HQQLinear in quantizer_hqq.py * revert to check HQQLinear in quantizer_hqq.py * update HqqConfig default params * make ci happy * make ci happy * revert to HQQLinear check in quantizer_hqq.py * check hqq_min version 0.2.0 * set axis=1 as default in quantization_config.py * validate_env with hqq>=0.2.0 version message * deprecated hqq kwargs message * make ci happy * remove run_expected_keys_check hack + bump to 0.2.1 min hqq version * fix unexpected_keys hqq update * add pre_quantized check * add update_expected_keys to base quantizerr * ci base.py fix? * ci base.py fix? * fix "quantization typo" src/transformers/utils/quantization_config.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * fix post merge --------- Co-authored-by: Marc Sun <marc@huggingface.co> Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
ArthurZucker
left a comment
ArthurZucker
left a comment
There was a problem hiding this comment.
Something went wrong with the rebasing / merging as you have unrelated changes!
| } | ||
|
|
||
|
|
||
| class GlmDecoderLayer(nn.Module): |
There was a problem hiding this comment.
this one looks fairly classic, I would have supposed you don't need the forward (unless the issue is with the name of layers?)
Yes, currently looking at it |
|
Superseed by #33823 |
4 participants
What does this PR do?
Adds GLM model.