README.md

---
license: mit
pipeline_tag: text-generation
library_name: transformers
language: [
    'en', 'am', 'ar', 'as', 'az', 'be', 'bg', 'bn', 'br', 'bs', 'ca', 'cs', 'cy', 'da', 'de', 'el',
    'eo', 'es', 'et', 'eu', 'fa', 'ff', 'fi', 'fr', 'fy', 'ga', 'gd', 'gl', 'gn', 'gu', 'ha', 'he',
    'hi', 'hr', 'ht', 'hu', 'hy', 'id', 'ig', 'is', 'it', 'ja', 'jv', 'ka', 'kk', 'km', 'kn', 'ko',
    'ku', 'ky', 'la', 'lg', 'li', 'ln', 'lo', 'lt', 'lv', 'mg', 'mk', 'ml', 'mn', 'mr', 'ms', 'my',
    'ne', 'nl', 'no', 'ns', 'om', 'or', 'pa', 'pl', 'ps', 'pt', 'qu', 'rm', 'ro', 'ru', 'sa', 'si',
    'sc', 'sd', 'sk', 'sl', 'so', 'sq', 'sr', 'ss', 'su', 'sv', 'sw', 'ta', 'te', 'th', 'tl', 'tn',
    'tr', 'ug', 'uk', 'ur', 'uz', 'vi', 'wo', 'xh', 'yi', 'yo', 'zu',
]
datasets:
# core - base
- ontocord/fineweb-permissive-multilingual-2m
- distily/c4_multilingual_1M
- data-silence/sumnews
- xu-song/cc100-samples
- badrex/llm-emoji-dataset
- fblgit/simple-math
- Gusarich/math-expressions-1m
- neuralwork/arxiver
- christopher/rosetta-code
- nampdn-ai/tiny-codes
- JeanKaddour/minipile
# core - instruct
- NousResearch/hermes-function-calling-v1
- simplescaling/s1K-1.1
# base - instruct
- mlabonne/open-perfectblend
- allenai/tulu-3-sft-mixture
- rombodawg/Everything_Instruct_Multilingual
# base - reason
- open-r1/OpenR1-Math-220k
- open-thoughts/OpenThoughts-114k
- cognitivecomputations/dolphin-r1
- simplescaling/s1K-1.1
tags:
- chat
- core
- base
- instruct
- reason
---

# tangled-alpha-0.3-core

![logo](./misc/logo.jpg)

```bash
time python -B prepare_core_datasets.py
```

```
i=0, min_len=0, max_len=1048576, block_size=2049, chunk_size=16392000, len(dataset)=3134311, len(dataset) * block_size=6422203239
Total number of tokens in the optimized dataset '../core-data-0-0-1048576-2049-8000' is 6422203239

i=1, min_len=2049, max_len=8193, block_size=8193, chunk_size=16386000, len(dataset)=179944, len(dataset) * block_size=1474281192
Total number of tokens in the optimized dataset '../core-data-1-2049-8193-8193-2000' is 1474281192

i=2, min_len=8193, max_len=1048577, block_size=32769, chunk_size=16384500, len(dataset)=48261, len(dataset) * block_size=1581464709
Total number of tokens in the optimized dataset '../core-data-2-8193-1048577-32769-500' is 1581464709
```

```bash
CUDA_VISIBLE_DEVICES=0 CUDA_LAUNCH_BLOCKING=0 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True litgpt pretrain --config pretrain-core-model-0.yaml
```

```
Seed set to 23
Time to instantiate model: 0.30 seconds.
Total parameters: 185,631,232
Verifying settings ...
Measured TFLOPs: 14094.64
Epoch 1 | iter 128 step 1 | loss train: 11.709, val: n/a | iter time: 341.75 ms (step) remaining time: 3 days, 20:04:36
Epoch 1 | iter 256 step 2 | loss train: 11.716, val: n/a | iter time: 287.55 ms (step) remaining time: 3 days, 3:29:34
Epoch 1 | iter 384 step 3 | loss train: 11.711, val: n/a | iter time: 290.88 ms (step) remaining time: 2 days, 22:16:53
Epoch 1 | iter 512 step 4 | loss train: 11.706, val: n/a | iter time: 291.81 ms (step) remaining time: 2 days, 19:34:34
Epoch 1 | iter 640 step 5 | loss train: 11.696, val: n/a | iter time: 291.37 ms (step) remaining time: 2 days, 17:59:17
Epoch 1 | iter 768 step 6 | loss train: 11.687, val: n/a | iter time: 290.50 ms (step) remaining time: 2 days, 16:55:49
Epoch 1 | iter 896 step 7 | loss train: 11.675, val: n/a | iter time: 291.08 ms (step) remaining time: 2 days, 16:10:38
Epoch 1 | iter 1024 step 8 | loss train: 11.660, val: n/a | iter time: 294.46 ms (step) remaining time: 2 days, 15:36:26
Epoch 1 | iter 1152 step 9 | loss train: 11.640, val: n/a | iter time: 292.26 ms (step) remaining time: 2 days, 15:09:28
Epoch 1 | iter 1280 step 10 | loss train: 11.626, val: n/a | iter time: 289.93 ms (step) remaining time: 2 days, 14:47:34
Epoch 1 | iter 1408 step 11 | loss train: 11.584, val: n/a | iter time: 292.15 ms (step) remaining time: 2 days, 14:29:19
Epoch 1 | iter 1536 step 12 | loss train: 11.526, val: n/a | iter time: 291.24 ms (step) remaining time: 2 days, 14:13:54
Epoch 1 | iter 1664 step 13 | loss train: 11.483, val: n/a | iter time: 291.11 ms (step) remaining time: 2 days, 14:00:48
Epoch 1 | iter 1792 step 14 | loss train: 11.430, val: n/a | iter time: 290.68 ms (step) remaining time: 2 days, 13:49:24
Epoch 1 | iter 1920 step 15 | loss train: 11.392, val: n/a | iter time: 290.37 ms (step) remaining time: 2 days, 13:39:22
Epoch 1 | iter 2048 step 16 | loss train: 11.326, val: n/a | iter time: 290.31 ms (step) remaining time: 2 days, 13:30:34
Epoch 1 | iter 2176 step 17 | loss train: 11.279, val: n/a | iter time: 290.33 ms (step) remaining time: 2 days, 13:22:34
Epoch 1 | iter 2304 step 18 | loss train: 11.222, val: n/a | iter time: 290.50 ms (step) remaining time: 2 days, 13:15:27
Epoch 1 | iter 2432 step 19 | loss train: 11.163, val: n/a | iter time: 290.39 ms (step) remaining time: 2 days, 13:09:11
Epoch 1 | iter 2560 step 20 | loss train: 11.094, val: n/a | iter time: 290.00 ms (step) remaining time: 2 days, 13:03:21
# ...
Epoch 1 | iter 782592 step 6114 | loss train: 3.080, val: 3.255 | iter time: 288.91 ms (step) remaining time: 0:06:14
Epoch 1 | iter 782720 step 6115 | loss train: 3.096, val: 3.255 | iter time: 289.11 ms (step) remaining time: 0:05:39
Epoch 1 | iter 782848 step 6116 | loss train: 2.977, val: 3.255 | iter time: 289.28 ms (step) remaining time: 0:05:04
Epoch 1 | iter 782976 step 6117 | loss train: 3.040, val: 3.255 | iter time: 289.24 ms (step) remaining time: 0:04:29
Epoch 1 | iter 783104 step 6118 | loss train: 3.062, val: 3.255 | iter time: 290.49 ms (step) remaining time: 0:03:54
Epoch 1 | iter 783232 step 6119 | loss train: 3.037, val: 3.255 | iter time: 289.91 ms (step) remaining time: 0:03:19
Epoch 1 | iter 783360 step 6120 | loss train: 3.028, val: 3.255 | iter time: 289.49 ms (step) remaining time: 0:02:44
Epoch 1 | iter 783488 step 6121 | loss train: 3.007, val: 3.255 | iter time: 289.81 ms (step) remaining time: 0:02:09
Epoch 2 | iter 783616 step 6122 | loss train: 3.007, val: 3.255 | iter time: 289.34 ms (step) remaining time: 0:01:34
Epoch 2 | iter 783744 step 6123 | loss train: 3.046, val: 3.255 | iter time: 288.52 ms (step) remaining time: 0:00:59
Epoch 2 | iter 783872 step 6124 | loss train: 3.140, val: 3.255 | iter time: 288.66 ms (step) remaining time: 0:00:24
Validating ...
Final evaluation | val loss: 3.254 | val ppl: 25.904
Saving checkpoint to '../out/pretrain-core-0/final/lit_model.pth'
----------------------------------------
| Performance
| - Total tokens  : 6,422,200,320
| - Training Time : 214857.29 s
| - Tok/sec       : 109674.70 tok/s
| ----------------------------------------
| Memory Usage
| - Memory Used   : 17.30 GB
----------------------------------------
```

Backup `wandb`:

```bash
mv wandb wandb-pretrain-core
```

Chat with model:

```bash
CUDA_VISIBLE_DEVICES=0 CUDA_LAUNCH_BLOCKING=0 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True litgpt chat ../out/pretrain-core-0/final
```

```bash
CUDA_VISIBLE_DEVICES=0 CUDA_LAUNCH_BLOCKING=0 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True time litgpt evaluate --tasks 'leaderboard' --out_dir '../evaluate/pretrain-core-0/leaderboard/' --batch_size 1 --dtype 'bfloat16' '../out/pretrain-core-0/final'
```

```
# ...
```