661 Commits

Author SHA1 Message Date
Sebastian Raschka
092b5b5429 topk comment 2024-08-20 20:44:15 -05:00
rasbt
f4e45a3f40 add note about duplicated cell 2024-08-19 21:04:18 -05:00
Sebastian Raschka
11e2f56af5 Note about MPS devices (#329) 2024-08-19 20:58:45 -05:00
Sebastian Raschka
a82169290e Note about MPS in ch06 and ch07 (#325) 2024-08-19 08:11:33 -05:00
Sebastian Raschka
0991c1ff24 Note about ch05 mps support (#324) 2024-08-19 07:40:24 -05:00
rasbt
1a962f3983 update lora experiments 2024-08-16 08:57:46 -05:00
rasbt
742525cb28 remove redundant indentation 2024-08-16 07:54:02 -05:00
TITC
70a7cddf9c typo fix (#323) 2024-08-16 06:57:30 -05:00
TITC
e1072bbcd6 typo fix (#321)
* typo fix

experiment 5 is the same size as experiment 1, both are 124 Million. and experiment 6 is 355M ~3x 124M.

* typo fix

all layer FT vs all layer FT, "slightly worse by 1.3% " indicates it's exp 5 vs exp 9

* exp 5 vs 9

* Update ch06/02_bonus_additional-experiments/README.md

---------

Co-authored-by: Sebastian Raschka <mail@sebastianraschka.com>
2024-08-15 08:45:34 -05:00
Sebastian Raschka
e4cfcc5b66 Update README.md 2024-08-15 08:22:47 -05:00
Daniel Kleine
bf1d8687f4 added std error bars (#320)
* added std error bars

* fixed changes

* Update on A100

---------

Co-authored-by: Sebastian Raschka <mail@sebastianraschka.com>
2024-08-13 20:57:41 -05:00
Sebastian Raschka
bb73efb5cf Consistency update in README.md 2024-08-13 07:33:10 -05:00
TITC
0b998dff97 track tokens seen in chapter5, track examples seen in chapter6 (#319) 2024-08-13 07:09:05 -05:00
rasbt
6533ce63c1 fix code cell ordering 2024-08-12 19:04:05 -05:00
Jeroen Van Goey
75abb61f13 Small typo fix (#313)
* typo fix

* Update ch03/02_bonus_efficient-multihead-attention/mha-implementations.ipynb

* Update ch03/02_bonus_efficient-multihead-attention/mha-implementations.ipynb

* Update ch03/02_bonus_efficient-multihead-attention/mha-implementations.ipynb

---------

Co-authored-by: Sebastian Raschka <mail@sebastianraschka.com>
2024-08-12 07:54:12 -05:00
Eric Thomson
b3d550bfd5 Adds .vscode folder to .gitignore (#314)
* added .vscode folder to .gitignore

* Update .gitignore

---------

Co-authored-by: Sebastian Raschka <mail@sebastianraschka.com>
2024-08-12 07:49:11 -05:00
Sebastian Raschka
f58c2b969b update attention benchmarks (#307) 2024-08-10 09:44:11 -05:00
Sebastian Raschka
8d79fb13b0 Update README.md 2024-08-10 07:54:51 -05:00
Daniel Kleine
c91999b9f4 fixed bash command (#305)
Co-authored-by: Sebastian Raschka <mail@sebastianraschka.com>
2024-08-09 21:29:04 -05:00
TITC
3067ed83dc remove all non-English texts and notice (#304)
* remove all non-English texts and notice

1. almost 18GB txt left after `is_english` filtered.
2. remove notice use gutenberg's strip_headers
3. after re-run get_data.py, seems all data are under `gutenberg/data/.mirror` folder.

* some improvements

* update readme

---------

Co-authored-by: rasbt <mail@sebastianraschka.com>
2024-08-09 17:09:14 -05:00
Sebastian Raschka
de277ebfe6 Update README.md 2024-08-08 07:50:45 -05:00
Sebastian Raschka
4601219c32 Update README.md 2024-08-08 07:47:31 -05:00
Sebastian Raschka
895512ebee Revert accidental edit 2024-08-06 19:54:34 -05:00
rasbt
7a5771932b note about logistic sigmoid 2024-08-06 19:48:30 -05:00
rasbt
9ef132bb64 note about logistic sigmoid 2024-08-06 19:48:06 -05:00
rasbt
2245f8d9c1 extend equation description 2024-08-06 19:46:50 -05:00
rasbt
a65e06ff99 add more explanations 2024-08-06 19:45:11 -05:00
Sebastian Raschka
b1a420c8ca Update README.md 2024-08-06 08:02:01 -05:00
TITC
7374d617b4 total training iters may equal to warmup_iters (#301)
total_training_iters=20, warmup_iters=20= len(train_loader) 4 multiply n_epochs 5, then ZeroDivisionError occurred.
```shell
Traceback (most recent call last):                                                                                                                                                                                                                                                                                              
  File "LLMs-from-scratch/ch05/05_bonus_hparam_tuning/hparam_search.py", line 191, in <module>                                             
    train_loss, val_loss = train_model(                                                                                                                                                                                                                                                                                         
                           ^^^^^^^^^^^^                                                                                                                         
  File "/mnt/raid1/docker/ai/LLMs-from-scratch/ch05/05_bonus_hparam_tuning/hparam_search.py", line 90, in train_model                                                                                                                                                                                                           
    progress = (global_step - warmup_iters) / (total_training_iters - warmup_iters)                                                                             
               ~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~                                                                                                                                                                                                                                             
ZeroDivisionError: division by zero 
```
2024-08-06 07:10:05 -05:00
Sebastian Raschka
192bdc3501 improve gradient accumulation (#300) 2024-08-05 18:27:20 -05:00
rasbt
089901db26 small figure update 2024-08-05 17:57:16 -05:00
Sebastian Raschka
b39234fc25 Update README.md 2024-08-05 17:47:06 -05:00
Daniel Kleine
dcdf04e3bd minor DPO fixes (#298)
* fixed issues, updated .gitignore

* added closing paren

* fixed CEL spelling

* fixed more minor issues

* Update ch07/01_main-chapter-code/ch07.ipynb

* Update ch07/04_preference-tuning-with-dpo/dpo-from-scratch.ipynb

* Update ch07/04_preference-tuning-with-dpo/dpo-from-scratch.ipynb

* Update ch07/04_preference-tuning-with-dpo/dpo-from-scratch.ipynb

---------

Co-authored-by: Sebastian Raschka <mail@sebastianraschka.com>
2024-08-05 08:40:46 -05:00
rasbt
6030071e3f update model path 2024-08-05 07:36:08 -05:00
SSebo
22681878a8 Update ch05.ipynb (#297)
typo
2024-08-05 07:12:27 -05:00
Sebastian Raschka
e4ed2c91cc Update README.md 2024-08-04 16:06:38 -05:00
rasbt
f302f5e8d5 improve latex rendering in dpo notebook 2024-08-04 09:19:59 -05:00
Sebastian Raschka
2c6cdb497f Update matplotlib tests on Windows (#295) 2024-08-04 09:18:19 -05:00
Sebastian Raschka
09dc080cf3 Direct Preference Optimization from scratch (#294) 2024-08-04 08:57:36 -05:00
Sebastian Raschka
3ea0798d44 Update README.md 2024-08-01 18:17:42 -05:00
rasbt
eabeab04d1 restructure into local and cloud setup 2024-07-31 06:59:04 -05:00
Sebastian Raschka
b4b2ac668c Add video tutorial 2024-07-30 06:57:46 -05:00
Sebastian Raschka
6c2d34b03d Update README.md 2024-07-30 06:55:41 -05:00
rasbt
358717870b add state_dict example 2024-07-28 14:15:32 -05:00
rasbt
ce33e706ba Fix 8-billion-parameter spelling 2024-07-28 10:48:56 -05:00
Daniel Kleine
60752e3b3a fixed typos and formatting (#291) 2024-07-28 10:04:33 -05:00
Sebastian Raschka
75a24c4897 Update README.md 2024-07-28 09:28:11 -05:00
Sebastian Raschka
011ee83ea7 Update README.md 2024-07-28 08:21:38 -05:00
Sebastian Raschka
1b86a61a2e buffer tutorial 2024-07-27 17:06:16 -05:00
Sebastian Raschka
f66c089f0b Test with PyTorch 2.0 and 2.4 (#290)
* Test with PyTorch 2.0 and 2.4

* Update basic-tests-old-pytorch.yml

* skip version cell
2024-07-27 15:09:02 -05:00