112 lines
6.3 KiB
Markdown
112 lines
6.3 KiB
Markdown
---
|
|
license: apache-2.0
|
|
datasets:
|
|
- BUT-FIT/BUT-LCC
|
|
- BUT-FIT/adult_content_classifier_dataset
|
|
language:
|
|
- cs
|
|
---
|
|
# Introduction
|
|
CSTinyLlama-1.2B is a Czech language model continously pretrained on 168b training tokens from English [TinyLLama-2.5T](https://huggingface.co/TinyLlama/TinyLlama-1.1B-intermediate-step-1195k-token-2.5T) model. Model was pretrained on ~67b token [Large Czech Collection](https://huggingface.co/datasets/BUT-FIT/BUT-LCC) using Czech tokenizer, obtained using our vocabulary swap method.
|
|
Training was done on [Karolina](https://www.it4i.cz/en) cluster.
|
|
|
|
# <span style="color:red">BUT LM Model Roster</span>
|
|
- [BUT-FIT/CSTinyLlama-1.2B](https://huggingface.co/BUT-FIT/CSTinyLlama-1.2B)
|
|
- [BUT-FIT/Czech-GPT-2-XL-133k](https://huggingface.co/BUT-FIT/Czech-GPT-2-XL-133k)
|
|
- [BUT-FIT/csmpt7b](https://huggingface.co/BUT-FIT/csmpt7b)
|
|
|
|
# Loss
|
|
Below we
|
|
- (i) demonstrate the convergence speed of released model (`TINYLLAMA1.2B_cztokenizer64k_align1.7k_tllama1.1B_C2048_lr1e-04_150k`, at 160k step).
|
|
- (ii) justify the contributions of our vocabulary swap method by comparing the swapped model with model trained from scratch (using same hyperparameters) `scratch_cztokenizer64k_tllama1.1B_C2048_lr1e-04_150k`. We swap 1.7K tokens in this run, similarly as for our other models (see [Czech-GPT-2-XL-133k](https://huggingface.co/BUT-FIT/Czech-GPT-2-XL-133k))
|
|
|
|
## Train Cross-Entropy
|
|
<img src="figures/tllama_train.png" width="900"/>
|
|
|
|
## Test Perplexity
|
|
<img src="figures/tllama_test.png" width="900"/>
|
|
|
|
## Distance in Steps For the Same Loss from Fine-Tuning vs Training from Scratch
|
|
<img src="figures/tllama_test_distance.png" width="900"/>
|
|
The distance |x1-x2| with same function value f1(x1)=f2(x2) grows with more steps. On convergence, it starts to rapidly increase (perhaps exponentially).
|
|
|
|
## Training parameters
|
|
Not mentioned parameters are the same as for [TinyLLama-2.5T](https://huggingface.co/TinyLlama/TinyLlama-1.1B-intermediate-step-1195k-token-2.5T).
|
|
|
|
| **Name** | **Value** | **Note** |
|
|
|----------------------------|---------------|----------------------------------------------------------------------------------------------|
|
|
| dataset_type | Concat | Sequences at the model's input were concatenated up to `$max_seq_len`, divided by EOS token. |
|
|
| tokenizer_size | 64k | |
|
|
| max_seq_len | 2048 | |
|
|
| batch_size | 512 | |
|
|
| learning_rate | 1.0e-4 | |
|
|
| optimizer | LionW | |
|
|
| optimizer_betas | 0.9/0.95 | |
|
|
| optimizer_weight_decay | 0 | |
|
|
| gradient_clipping_max_norm | 1.0 | |
|
|
| attn_impl | flash2 | |
|
|
| fsdp | SHARD_GRAD_OP | (optimized for A100 40GB GPUs) |
|
|
| precision | bf16 | |
|
|
| scheduler | cosine | |
|
|
| scheduler_warmup | 100 steps | |
|
|
| scheduler_steps | 200,000 | |
|
|
| scheduler_alpha | 0.1 | So LR on last step is 0.1*(vanilla LR) |
|
|
|
|
|
|
## Usage
|
|
```python
|
|
import torch
|
|
import transformers
|
|
from transformers import pipeline
|
|
|
|
name = 'BUT-FIT/CSTinyLlama-1.2B'
|
|
|
|
config = transformers.AutoConfig.from_pretrained(name, trust_remote_code=True)
|
|
model = transformers.AutoModelForCausalLM.from_pretrained(
|
|
name,
|
|
config=config,
|
|
trust_remote_code=True
|
|
)
|
|
|
|
tokenizer = transformers.AutoTokenizer.from_pretrained(name, trust_remote_code=True)
|
|
|
|
pipe = pipeline('text-generation', model=model, tokenizer=tokenizer, device='cuda:0')
|
|
|
|
with torch.autocast('cuda', dtype=torch.bfloat16):
|
|
print(
|
|
pipe('Nejznámějším českým spisovatelem ',
|
|
max_new_tokens=100,
|
|
top_p=0.95,
|
|
repetition_penalty=1.0,
|
|
do_sample=True,
|
|
use_cache=True))
|
|
```
|
|
# Training Data
|
|
We release most (95.79%) of our training data corpus as [BUT-Large Czech Collection](https://huggingface.co/datasets/BUT-FIT/but_lcc).
|
|
|
|
|
|
## Getting in Touch
|
|
For further questions, email to `martin.fajcik@vut.cz`.
|
|
|
|
# Disclaimer
|
|
This is a probabilistic model, it can output stochastic information. Authors are not responsible for the model outputs. Use at your own risk.
|
|
|
|
# Acknowledgement
|
|
This work was supported by NAKI III program of Ministry of Culture Czech Republic, project semANT ---
|
|
"Sémantický průzkumník textového kulturního dědictví" grant no. `DH23P03OVV060` and
|
|
by the Ministry of Education, Youth and Sports of the Czech Republic through the e-INFRA CZ (ID:`90254`).
|
|
|
|
# Citation
|
|
```bibtex
|
|
@article{benczechmark,
|
|
author = {Martin Fajčík, Martin Dočekal, Jan Doležal, Karel Beneš, Michal Hradiš},
|
|
title = {BenCzechMark: Machine Language Understanding Benchmark for Czech Language},
|
|
journal = {arXiv preprint arXiv:insert-arxiv-number-here},
|
|
year = {2024},
|
|
month = {March},
|
|
eprint = {insert-arxiv-number-here},
|
|
archivePrefix = {arXiv},
|
|
primaryClass = {cs.CL},
|
|
}
|
|
|