Skip to content

Commit d7ea4d2

Browse files
tysam-codetysam-code
committed
Initial commit for gpt-hlb
0 parents  commit d7ea4d2

File tree

5 files changed

+776
-0
lines changed

5 files changed

+776
-0
lines changed

‎CITATION.cff

+9
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,9 @@
1+
cff-version: 1.2.0
2+
message: "If you need to cite this codebase for any reason, please do so as below."
3+
authors:
4+
- family-names: "Balsam"
5+
given-names: "Tysam&"
6+
title: "hlb-gpt"
7+
version: 0.0.0
8+
date-released: 2023-03-05
9+
url: "https://github.com/tysam-code/hlb-gpt"

‎LICENSE.md

+201
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,201 @@
1+
Apache License
2+
Version 2.0, January 2004
3+
http://www.apache.org/licenses/
4+
5+
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
6+
7+
1. Definitions.
8+
9+
"License" shall mean the terms and conditions for use, reproduction,
10+
and distribution as defined by Sections 1 through 9 of this document.
11+
12+
"Licensor" shall mean the copyright owner or entity authorized by
13+
the copyright owner that is granting the License.
14+
15+
"Legal Entity" shall mean the union of the acting entity and all
16+
other entities that control, are controlled by, or are under common
17+
control with that entity. For the purposes of this definition,
18+
"control" means (i) the power, direct or indirect, to cause the
19+
direction or management of such entity, whether by contract or
20+
otherwise, or (ii) ownership of fifty percent (50%) or more of the
21+
outstanding shares, or (iii) beneficial ownership of such entity.
22+
23+
"You" (or "Your") shall mean an individual or Legal Entity
24+
exercising permissions granted by this License.
25+
26+
"Source" form shall mean the preferred form for making modifications,
27+
including but not limited to software source code, documentation
28+
source, and configuration files.
29+
30+
"Object" form shall mean any form resulting from mechanical
31+
transformation or translation of a Source form, including but
32+
not limited to compiled object code, generated documentation,
33+
and conversions to other media types.
34+
35+
"Work" shall mean the work of authorship, whether in Source or
36+
Object form, made available under the License, as indicated by a
37+
copyright notice that is included in or attached to the work
38+
(an example is provided in the Appendix below).
39+
40+
"Derivative Works" shall mean any work, whether in Source or Object
41+
form, that is based on (or derived from) the Work and for which the
42+
editorial revisions, annotations, elaborations, or other modifications
43+
represent, as a whole, an original work of authorship. For the purposes
44+
of this License, Derivative Works shall not include works that remain
45+
separable from, or merely link (or bind by name) to the interfaces of,
46+
the Work and Derivative Works thereof.
47+
48+
"Contribution" shall mean any work of authorship, including
49+
the original version of the Work and any modifications or additions
50+
to that Work or Derivative Works thereof, that is intentionally
51+
submitted to Licensor for inclusion in the Work by the copyright owner
52+
or by an individual or Legal Entity authorized to submit on behalf of
53+
the copyright owner. For the purposes of this definition, "submitted"
54+
means any form of electronic, verbal, or written communication sent
55+
to the Licensor or its representatives, including but not limited to
56+
communication on electronic mailing lists, source code control systems,
57+
and issue tracking systems that are managed by, or on behalf of, the
58+
Licensor for the purpose of discussing and improving the Work, but
59+
excluding communication that is conspicuously marked or otherwise
60+
designated in writing by the copyright owner as "Not a Contribution."
61+
62+
"Contributor" shall mean Licensor and any individual or Legal Entity
63+
on behalf of whom a Contribution has been received by Licensor and
64+
subsequently incorporated within the Work.
65+
66+
2. Grant of Copyright License. Subject to the terms and conditions of
67+
this License, each Contributor hereby grants to You a perpetual,
68+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
69+
copyright license to reproduce, prepare Derivative Works of,
70+
publicly display, publicly perform, sublicense, and distribute the
71+
Work and such Derivative Works in Source or Object form.
72+
73+
3. Grant of Patent License. Subject to the terms and conditions of
74+
this License, each Contributor hereby grants to You a perpetual,
75+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
76+
(except as stated in this section) patent license to make, have made,
77+
use, offer to sell, sell, import, and otherwise transfer the Work,
78+
where such license applies only to those patent claims licensable
79+
by such Contributor that are necessarily infringed by their
80+
Contribution(s) alone or by combination of their Contribution(s)
81+
with the Work to which such Contribution(s) was submitted. If You
82+
institute patent litigation against any entity (including a
83+
cross-claim or counterclaim in a lawsuit) alleging that the Work
84+
or a Contribution incorporated within the Work constitutes direct
85+
or contributory patent infringement, then any patent licenses
86+
granted to You under this License for that Work shall terminate
87+
as of the date such litigation is filed.
88+
89+
4. Redistribution. You may reproduce and distribute copies of the
90+
Work or Derivative Works thereof in any medium, with or without
91+
modifications, and in Source or Object form, provided that You
92+
meet the following conditions:
93+
94+
(a) You must give any other recipients of the Work or
95+
Derivative Works a copy of this License; and
96+
97+
(b) You must cause any modified files to carry prominent notices
98+
stating that You changed the files; and
99+
100+
(c) You must retain, in the Source form of any Derivative Works
101+
that You distribute, all copyright, patent, trademark, and
102+
attribution notices from the Source form of the Work,
103+
excluding those notices that do not pertain to any part of
104+
the Derivative Works; and
105+
106+
(d) If the Work includes a "NOTICE" text file as part of its
107+
distribution, then any Derivative Works that You distribute must
108+
include a readable copy of the attribution notices contained
109+
within such NOTICE file, excluding those notices that do not
110+
pertain to any part of the Derivative Works, in at least one
111+
of the following places: within a NOTICE text file distributed
112+
as part of the Derivative Works; within the Source form or
113+
documentation, if provided along with the Derivative Works; or,
114+
within a display generated by the Derivative Works, if and
115+
wherever such third-party notices normally appear. The contents
116+
of the NOTICE file are for informational purposes only and
117+
do not modify the License. You may add Your own attribution
118+
notices within Derivative Works that You distribute, alongside
119+
or as an addendum to the NOTICE text from the Work, provided
120+
that such additional attribution notices cannot be construed
121+
as modifying the License.
122+
123+
You may add Your own copyright statement to Your modifications and
124+
may provide additional or different license terms and conditions
125+
for use, reproduction, or distribution of Your modifications, or
126+
for any such Derivative Works as a whole, provided Your use,
127+
reproduction, and distribution of the Work otherwise complies with
128+
the conditions stated in this License.
129+
130+
5. Submission of Contributions. Unless You explicitly state otherwise,
131+
any Contribution intentionally submitted for inclusion in the Work
132+
by You to the Licensor shall be under the terms and conditions of
133+
this License, without any additional terms or conditions.
134+
Notwithstanding the above, nothing herein shall supersede or modify
135+
the terms of any separate license agreement you may have executed
136+
with Licensor regarding such Contributions.
137+
138+
6. Trademarks. This License does not grant permission to use the trade
139+
names, trademarks, service marks, or product names of the Licensor,
140+
except as required for reasonable and customary use in describing the
141+
origin of the Work and reproducing the content of the NOTICE file.
142+
143+
7. Disclaimer of Warranty. Unless required by applicable law or
144+
agreed to in writing, Licensor provides the Work (and each
145+
Contributor provides its Contributions) on an "AS IS" BASIS,
146+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
147+
implied, including, without limitation, any warranties or conditions
148+
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
149+
PARTICULAR PURPOSE. You are solely responsible for determining the
150+
appropriateness of using or redistributing the Work and assume any
151+
risks associated with Your exercise of permissions under this License.
152+
153+
8. Limitation of Liability. In no event and under no legal theory,
154+
whether in tort (including negligence), contract, or otherwise,
155+
unless required by applicable law (such as deliberate and grossly
156+
negligent acts) or agreed to in writing, shall any Contributor be
157+
liable to You for damages, including any direct, indirect, special,
158+
incidental, or consequential damages of any character arising as a
159+
result of this License or out of the use or inability to use the
160+
Work (including but not limited to damages for loss of goodwill,
161+
work stoppage, computer failure or malfunction, or any and all
162+
other commercial damages or losses), even if such Contributor
163+
has been advised of the possibility of such damages.
164+
165+
9. Accepting Warranty or Additional Liability. While redistributing
166+
the Work or Derivative Works thereof, You may choose to offer,
167+
and charge a fee for, acceptance of support, warranty, indemnity,
168+
or other liability obligations and/or rights consistent with this
169+
License. However, in accepting such obligations, You may act only
170+
on Your own behalf and on Your sole responsibility, not on behalf
171+
of any other Contributor, and only if You agree to indemnify,
172+
defend, and hold each Contributor harmless for any liability
173+
incurred by, or claims asserted against, such Contributor by reason
174+
of your accepting any such warranty or additional liability.
175+
176+
END OF TERMS AND CONDITIONS
177+
178+
APPENDIX: How to apply the Apache License to your work.
179+
180+
To apply the Apache License to your work, attach the following
181+
boilerplate notice, with the fields enclosed by brackets "[]"
182+
replaced with your own identifying information. (Don't include
183+
the brackets!) The text should be enclosed in the appropriate
184+
comment syntax for the file format. We also recommend that a
185+
file or class name and description of purpose be included on the
186+
same "printed page" as the copyright notice for easier
187+
identification within third-party archives.
188+
189+
Copyright [yyyy] [name of copyright owner]
190+
191+
Licensed under the Apache License, Version 2.0 (the "License");
192+
you may not use this file except in compliance with the License.
193+
You may obtain a copy of the License at
194+
195+
http://www.apache.org/licenses/LICENSE-2.0
196+
197+
Unless required by applicable law or agreed to in writing, software
198+
distributed under the License is distributed on an "AS IS" BASIS,
199+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
200+
See the License for the specific language governing permissions and
201+
limitations under the License.

‎README.md

+72
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,72 @@
1+
[![Twitter URL](https://img.shields.io/twitter/url/https/twitter.com/hi_tysam.svg?style=social&label=Follow%20%40TySam_And)](https://twitter.com/hi_tysam) [![Support me on Patreon](https://img.shields.io/endpoint.svg?url=https%3A%2F%2Fshieldsio-patreon.vercel.app%2Fapi%3Fusername%3Dtysam%26type%3Dpatrons%26suffix%3Dsponsors&style=flat)](https://patreon.com/tysam)
2+
3+
## hlb-GPT
4+
Welcome to the hyperlightspeedbench-gpt (hlb-gpt) repo! :D
5+
6+
### How to Run
7+
8+
9+
`git clone https://github.com/tysam-code/hlb-gpt && cd hlb-gpt && python -m pip install -r requirements.txt && python main.py`
10+
11+
12+
If you're curious, this code is generally Colab friendly (in fact -- most of this was developed in Colab!). Just be sure to uncomment the reset block at the top of the code.
13+
14+
15+
### Main
16+
17+
Goals:
18+
* minimalistic
19+
* hackable
20+
* beginner-friendly
21+
* torch- and python-idiomatic
22+
* reference implementation for performant applied ML
23+
* few external dependencies (currently only torch, torchvision, and tiktoken)
24+
* eventual world-record-speed single-GPU training times on at least one LLM benchmark (currently via the A100! :D)
25+
26+
27+
This code implements a fast-training language model baseline that achieves under ~3.8 val loss (44.7 perplexity) on WikiText-103 in just over 6 minutes. It is currently a distilled, feature-pruned, relatively faithful reimplementation of a basic GPT language model as defined in Karpathy's excellent [nanoGPT](https://github.com/karpathy/nanoGPT) repository**. The only differences in this implentation that I know of are different function calls for the attention operation, the full-accuracy Pytorch GELU function, and the native PyTorch OneCycle scheduler. There are also different default hyperparameters specific to these short runs. For the rationale behind the 3.8 val loss target, please see the '[Why 3.8 Val Loss?](#why-3.8-val-loss?)' section.
28+
29+
30+
This is a very focused implementation which attempts to maximize code understandability and minimize code length. At the same time, it aims to be very hackable to let people test out new ideas and get back initial results rapidly. We also want to keep this code as accessible as possible to a wide range of users and usecases -- mainly through layout, helpful code comments, and simplicity. We also only target one piece of hardware -- the A100 currently -- but attempt to maintain accessibility by providing options for people with less GPU memory. As as result of all of this, this means that this implementation really doesn't have much in the way of fancy features. It downloads and loads the data, creates the network, runs the training loop and that's about it -- if you want anything on top of that, it should be easily to implement with how modular the code is. That said, feel free to open an issue if there is something critical that I've missed!
31+
32+
33+
Finally, this code is meant to be fast -- as fast as possible. Please keep an eye out for further training speedups in future updates, since this is just the baseline after all. This code is in a single file and extremely flat, but is not as durable for long-term production-level bug maintenance. You're meant to check out a fresh repo whenever you have a new idea. Part of the recommended workflow at scale is that if you're at an organization that needs a modified 'base repo', to modify this repo and use that as your new base internally. I oftentimes use a branching tree structure several repos deep in my work and I find it to be a great way to rapidly explore/context switch/roll back between different problem-solving domains. It's also another reason why I keep the base repo so simple.
34+
35+
36+
Speaking of which -- I believe that this repo is excellent for rapid idea exploration -- almost everywhere in the pipeline is exposed and built to be hackable/user-friendly for very fast, iterated research. I hope you enjoy using it! :D Please let me know if you have any feedback. I hope to continue publishing updates to this in the future, so your support and thoughts on what worked for you/what didn't is encouraged. Share this repo with someone you know that might like it! I can't wait to see what you come up with. :) :D
37+
38+
39+
Feel free to check out my [Patreon](https://www.patreon.com/user/posts?u=83632131) if you like what I'm doing here and want to support me in my work! Any funding for GPU hours is greatly appreciated! :)))) Additionally, if you want me to work up to a part-time amount of hours with you via short contract work or otherwise, feel free to reach out to me at [email protected]. I'd love to hear from you.
40+
41+
42+
43+
** I also recommend checking out the [GPT video](https://youtu.be/kCc8FmEb1nY) from his excellent zero to hero series if you are generally unfamiliar with large language models (LLMs) :) :D
44+
45+
### Known Bugs / Potential Problem Areas
46+
47+
The Colab-specific code is commented out at the top, and some of the model weight initialization and flops/mfu/etc calculations might require you to update them manually if you are making significant changes to the network.
48+
49+
### Why 3.8 Val Loss?
50+
51+
Basically -- here we have very similar reasoning to [hlb-CIFAR10](https://github.com/tysam-code/hlb-CIFAR10#why-a-convnet-still-why-cifar10-arent-transformers-the-new-thing-now) -- that is, the information we gain from an experiment (for the most part) should follow the same trend(s) as longer-running or larger scale experiments, albeit with some more noise. This isn't always true, but I've personally found it to hold true more than most people generally seem to expect it to. It does vary on a case-by case basis, but even for the edge cases, there is a good short-term analog that we can use. On the opposite hand, if we do longer/larger runs for most of our experiments, we can lose a lot of time, brainpower, unbiased experimenting, and the cost of context switching (!!!). Really short experiment cycles allow us to run more experiments and cover more exploratory ground, which gives us a broader exposure to the general shape of the problem than slow, methodical runs. It also allows us to reduce our bias and introduce some variance to the research process, because we don't always have to plan as hard before each run due to how short each experiment is. This introduces some noise into the experimentation process, which I've found to be a surprisingly important component for overcoming human bias and leading to new discoveries.
52+
53+
54+
There's a lot more to it than that, but especially in the realm of generative pretraining -- for the 'real world usecase', in the long runs, we're ideally/hopefully going to only be seeing each token with the full network only _once_. This means that for this problem, we basically have an open field where we just have to (hopefully) maximize the rate of information acquisition into the network as fast as we possibly can. A val loss of 3.8 is right around where the training curves start flattening out during early training, but it still requires the network to learn some temporal and topical dependencies to get to that point. Values like 3.6 or even 3.4 can work well, but they seem to be in the realm of diminishing returns, especially for initial exploration. If we get fast enough and we find that we should move the marker for some reason at some point in the future, then that's certainly possible! But for now, 3.8 (roughly) seems to be a good enough signal but without too much compromise in training time. For training taking just over ~6 minutes or so, the output from the networks is not too bad (provided the temperature is low enough)!
55+
56+
57+
One other bonus would be that for where we're at in the training process, a val loss of 3.8 I believe is just slightly over 1 single training pass over the data -- so hopefully with some optimization we can get there with only 1 training pass over the data, better simulating what we'd be doing with a larger dataset.
58+
59+
### Citation
60+
61+
If you use this work in your research, please cite
62+
`@software{Balsam_hlb-gpt_2023,
63+
author={Balsam, Tysam&},
64+
month={3},
65+
title={{hlb-gpt}},
66+
url={https://github.com/tysam-code/hlb-gpt},
67+
version = {0.0.0},
68+
year = {2023}}`
69+
70+
### Bugs & Etc.
71+
72+
If you find a bug, open an issue! L:D If you have a success story, let me know! It helps me understand what works and doesn't more than you might expect -- if I know how this is specifically helping people, that can help me further improve as a developer, as I can keep that in mind when developing other software for people in the future. :D :)

0 commit comments

Comments
 (0)