feat: medusa v2 #1734
Merged
feat: medusa v2 #1734
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
Narsil
reviewed
Apr 12, 2024
| super().__init__() | ||
| self.blocks = torch.nn.ModuleList( | ||
| [ | ||
| ResBlock(config, prefix=f"{prefix}.{i}", weights=weights) | ||
| for i in range(config["medusa_num_layers"]) | ||
| for i in range(medusa_config["medusa_num_layers"]) |
There was a problem hiding this comment.
Which should probably take that information from speculate instead (to avoid loading layers which we're not going to use)
We can do in another PR
|
|
||
| self.act = torch.nn.SiLU() | ||
|
|
||
| self.lm_head = TensorParallelHead.load(config, prefix, weights) |
There was a problem hiding this comment.
I thought what I had done was cleverer by passing the lm_head directly (which avoids a reload).
Since we're using sharding here Weights will actually load the LM head n times with current code. (iirc)
| medusa = MedusaModel(config, weights) | ||
| lm_head = None | ||
| try: | ||
| medusa = MedusaHeadV2(config, prefix, weights) |
There was a problem hiding this comment.
Shouldn't it be the reverse ? MedusaV2 will always load properly (since it's only a subset of V1) ?
| @@ -467,46 +467,159 @@ def forward(self, x): | |||
| return x | |||
|
|
|||
|
|
|||
| class SpeculativeHead(nn.Module): | |||
| class MedusaHeadV1(nn.Module): | |||
There was a problem hiding this comment.
Maybe it's time we put them into their own file (both in the same file is ok I think)?
Narsil
previously approved these changes
Apr 12, 2024
kdamaszk
pushed a commit
to kdamaszk/tgi-gaudi
that referenced
this pull request
Apr 29, 2024
Nilabhra
pushed a commit
to TII-AI-Research-Center/text-generation-inference
that referenced
this pull request
May 14, 2024
|
It would be helpful to document it. |
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.
No description provided.