File size: 1,166 Bytes
7b62092
 
 
 
 
 
 
28bf242
0f6bed2
28bf242
b53b44b
28bf242
 
 
 
284da44
 
 
28bf242
 
b53b44b
eea1bce
28bf242
 
 
b53b44b
28bf242
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b53b44b
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
---
license: llama2
language:
- ru
metrics:
- accuracy
---

# ruadapt_llama2_7b_v0.1

This model is a fine-tuned (embeddings, lm head) version of TheBloke/Llama-2-7B-fp16 on the Russian dataset (33GB).
It achieves the following results on the evaluation set:
- Loss: 2.7569
- Accuracy: 0.4617

Instruct version: 
https://huggingface.co/rccmsu/ruadapt_saiga2_7b_v0.1

## Model description

Russian adaptation of LLaMa-2-7B by replacing the tokenizer.
Paper: Tikhomirov M., Chernyshev D. Impact of Tokenization on LLaMa Russian Adaptation //arXiv preprint arXiv:2312.02598. – 2023.

## Intended uses & limitations

LLAMA 2 COMMUNITY LICENSE AGREEMENT	

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 6
- eval_batch_size: 6
- seed: 42
- distributed_type: multi-GPU
- num_devices: 16
- gradient_accumulation_steps: 2
- total_train_batch_size: 192
- total_eval_batch_size: 96
- optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-05
- lr_scheduler_type: linear
- num_epochs: 2.0

### Framework versions

- Transformers 4.34.0
- Pytorch 2.0.1+cu118
- Datasets 2.14.5
- Tokenizers 0.14.1