Add Typing to Llama Training #9
Open
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
|
Alright, will have to work on this some more as the checks indicate. |
|
Now it might go through. If someone has time, I recommend double checking the jaxtyping hints though, as I was not entirely sure everywhere. |
danbraunai
requested changes
Aug 8, 2024
simple_stories_train/train_llama.py
Outdated
Comment on lines
18
to
41
|
|
||
| This implementation is based on | ||
| - llm.c, licensed under MIT ((c) 2024 Andrei Karpathy) and | ||
| - TransformerLens, licensed under MIT ((c) 2022 TransformerLensOrg). | ||
|
|
||
|
|
||
| MIT License: | ||
| Permission is hereby granted, free of charge, to any person obtaining a copy | ||
| of this software and associated documentation files (the "Software"), to deal | ||
| in the Software without restriction, including without limitation the rights | ||
| to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | ||
| copies of the Software, and to permit persons to whom the Software is | ||
| furnished to do so, subject to the following conditions: | ||
|
|
||
| The above copyright notice and this permission notice shall be included in all | ||
| copies or substantial portions of the Software. | ||
|
|
||
| THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | ||
| IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | ||
| FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE | ||
| AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | ||
| LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | ||
| OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | ||
| SOFTWARE. |
There was a problem hiding this comment.
I think this was accidentally deleted?
simple_stories_train/train_llama.py
Outdated
| rotary_dim: int = 768 // 12 # i.e. same as d_head | ||
| rotary_base: int = 10000 | ||
| n_ctx: int = 1024 | ||
| n_key_value_heads: int = ( | ||
| 12 // 4 | ||
| ) # Note that llama 3.1 n_key_value_heads does not scale with n_heads | ||
| use_grouped_query_attention: bool = True |
simple_stories_train/train_llama.py
Outdated
Comment on lines
77
to
78
| self.kv_attn = nn.Linear(config.n_embd, 2 * config.n_embd // self.repeat_kv_heads) | ||
| self.q_attn = nn.Linear(config.n_embd, config.n_embd) |
There was a problem hiding this comment.
The bias argument has disappeared here, and for other attributes down below too.
simple_stories_train/train_llama.py
Outdated
| self.c_proj = nn.Linear(config.n_embd, config.n_embd, bias=config.attn_bias) | ||
| self.c_proj.LLMC_RESIDUAL_SCALE_FLAG = 1 | ||
| self.c_proj = nn.Linear(config.n_embd, config.n_embd) | ||
| self.LLMC_RESIDUAL_SCALE_FLAG = 1 |
There was a problem hiding this comment.
Why this change? Seems like you want to keep it on the c_proj
simple_stories_train/train_llama.py
Outdated
| @@ -576,7 +587,7 @@ def __init__(self, filename_pattern, B, T, process_rank, num_processes): | |||
| print0(f"DataLoader: total number of tokens: {ntok_total:,} across {len(self.files)} files") | |||
|
|
|||
| # kick things off | |||
| self.current_shard = None | |||
| self.current_shard = -1 | |||
There was a problem hiding this comment.
How come this change was made?
simple_stories_train/train_llama.py
Outdated
Comment on lines
621
to
626
| # ----------------------------------------------------------------------------- | ||
| # Python -> C bridge utilities for saving params/grads/activations to .bin files | ||
|
|
||
|
|
||
| def write_fp32(tensor: torch.Tensor, file: BufferedWriter): | ||
| t = tensor.detach().cpu().to(torch.float32) |
There was a problem hiding this comment.
I think we deleted all the below from main. I'm guessing you hadn't pulled that latest main onto your branch?
There was a problem hiding this comment.
Yes, looks like it.
simple_stories_train/train_llama.py
Outdated
Comment on lines
811
to
812
| import argparse | ||
| import time |
simple_stories_train/train_llama.py
Outdated
Comment on lines
1060
to
1063
| # ------------------------------------------------------------------------- | ||
| # PyTorch -> C bridge: save some weights and state for C to load later as reference | ||
|
|
||
| # do one forward pass to generate ground truth for our C tests |
There was a problem hiding this comment.
I think we deleted all this too
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.
Description
Added Jax and regular typing to
llama_train.pyRelated Issue
Should close #4.
How Has This Been Tested?
Executing the script in a Colab instance.