Merging DiVA to Levanter Main #779
Open
+412
−8
Conversation
This file contains 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
|
I am currently annoyed by how we initialize models and this seems fine enough (cf #780 ) so I don't have a super strong feeling right now on it. You could look to how we do Lora if you want, but that's a bit of a different case. |
dlwh
approved these changes
Oct 30, 2024
| lev_model: DivaModel = super().load_pretrained( | ||
| DivaModel, ref, config, axis_mapping, resize_vocab_to_match_tokenizer, dtype | ||
| ) # type: ignore[assignment] | ||
| llm: Union[LlamaLMHeadModel | MistralLMHeadModel | GemmaLMHeadModel] = HFCheckpointConverter( |
There was a problem hiding this comment.
Do you need this Union around your 3.10 union?
| elif "gemma" in model_id: | ||
| config = GemmaConfig.from_hf_config(hf_config) | ||
| elif "mistral" in model_id: | ||
| config = MistralConfig.from_hf_config(hf_config) |
There was a problem hiding this comment.
should you raise a better error if it's none of these?
| return config | ||
|
|
||
|
|
||
| def get_prefix(tokenizer_ref): |
| return prefix_tok, suffix_tok | ||
|
|
||
|
|
||
| @LmConfig.register_subclass("diva") |
| init_from_submodel: bool = True | ||
|
|
||
| # Connector Config | ||
| pre_audio_prompt = property(lambda self: get_prefix(self.reference_decoder)[0]) |
| ) | ||
|
|
||
|
|
||
| class DivaModel(eqx.Module, ModelWithHfSerializationMixin[DivaConfig]): |
|
|
||
| # Convert to Virtual LLM Tokens | ||
| virt_whisper_tokens = self.connector.transformer( | ||
| (self.query_tokens + self.query_position_embeds).broadcast_axis(OtherAxes), |
There was a problem hiding this comment.
hrm i wouldn't think this should be necessary
| text[ | ||
| { | ||
| "batch": hax.arange(Batch), | ||
| "position": (hax.sum(text_tokens == pad_token_id, "position") * -1) - 1, |
| kl_proxy_loss = hax.dot(diff_distill, diff_distill, axis="embed") ** 0.5 | ||
|
|
||
| # Compute Contrastive Loss on Input | ||
| # Correct for Normal Autoregressive Loss Mask |
There was a problem hiding this comment.
do you want to check that attn mask is causal or eh?
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.
Cleaned up version of my code for the Distilled Voice Assistant models that I trained using a fork of Levanter!
@dlwh Main thing I want to check in with you here is what the appropriate design pattern you think would make sense for initializing the model weights from multiple other pretrained models would be! What I've done here is much cleaner than what I did originally for the paper, but still feels a bit messy.
Testing Procedure for the correctness of this training code:
I trained a new DiVA model with this updated code and Llama 3.2 1B using the config in
diva_flash.yaml.Training Log is here: https://wandb.ai/i18nlp/levanter/runs/jnxp463y?nw=nwuserheld
Resulting model is on HF in PyTorch form here: https://huggingface.co/WillHeld/DiVA-llama-3.2-1b
Demo which confirmed the result is ~reasonable here for now: https://b3f161194b514a990f.gradio.live/