RefalMachine
commited on
Commit
•
ec34b94
1
Parent(s):
ee1059f
Update README.md
Browse files
README.md
CHANGED
@@ -10,19 +10,21 @@ base_model:
|
|
10 |
- RefalMachine/ruadapt_qwen2.5_3B_ext_u48_full_lr5e4_peft_mlp_32_32_bs256
|
11 |
---
|
12 |
|
13 |
-
|
14 |
|
15 |
Instruction-tuned version of RefalMachine/ruadapt_qwen2.5_3B_ext_u48_full_lr5e4_peft_mlp_32_32_bs256 with extended tokenizer after LEP (Learned Embedding Propagation, paper will be soon) procedure.
|
16 |
|
17 |
Thanks to the extended tokenizer, the model works more efficiently with the Russian language (up to 60% speed up compared to Qwen-2.5-3B-Instruct in terms of characters)
|
18 |
|
19 |
-
|
|
|
|
|
20 |
|
21 |
#### Результаты на Ru-Arena-General
|
22 |
|
23 |
В качестве референсых ответов, с которыми сравниваются модели выступают ответы от gpt-3.5-turbo-0125, поэтому она имеет винрейт 50%.
|
24 |
|
25 |
-
|
26 |
|
27 |
| Model Name | Winrate | 95% CI | Average # Tokens |
|
28 |
|--------------------------------------------------|--------|--------------------|------------------|
|
@@ -41,7 +43,15 @@ Thanks to the extended tokenizer, the model works more efficiently with the Russ
|
|
41 |
| c4ai-command-r-v01 | 49.0 | (-1.7, 2.2) | 529 |
|
42 |
| meta-llama-3.1-8b-instruct | 43.1 | (-2.8, 2.3) | 628 |
|
43 |
|
44 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
45 |
|
46 |
Tikhomirov M., Chernyshev D. Facilitating large language model Russian adaptation with Learned Embedding Propagation // 2024 (will be soon)
|
47 |
|
|
|
10 |
- RefalMachine/ruadapt_qwen2.5_3B_ext_u48_full_lr5e4_peft_mlp_32_32_bs256
|
11 |
---
|
12 |
|
13 |
+
## Model description
|
14 |
|
15 |
Instruction-tuned version of RefalMachine/ruadapt_qwen2.5_3B_ext_u48_full_lr5e4_peft_mlp_32_32_bs256 with extended tokenizer after LEP (Learned Embedding Propagation, paper will be soon) procedure.
|
16 |
|
17 |
Thanks to the extended tokenizer, the model works more efficiently with the Russian language (up to 60% speed up compared to Qwen-2.5-3B-Instruct in terms of characters)
|
18 |
|
19 |
+
## Метрики и оценка качества
|
20 |
+
|
21 |
+
Модель была оценена на Ru-Arena-General, MERA, llmtf_open
|
22 |
|
23 |
#### Результаты на Ru-Arena-General
|
24 |
|
25 |
В качестве референсых ответов, с которыми сравниваются модели выступают ответы от gpt-3.5-turbo-0125, поэтому она имеет винрейт 50%.
|
26 |
|
27 |
+
Приведена лишь часть лидерборда, подробнее смотрите в репозитории бенчмарка (https://huggingface.co/spaces/Vikhrmodels/arenahardlb).
|
28 |
|
29 |
| Model Name | Winrate | 95% CI | Average # Tokens |
|
30 |
|--------------------------------------------------|--------|--------------------|------------------|
|
|
|
43 |
| c4ai-command-r-v01 | 49.0 | (-1.7, 2.2) | 529 |
|
44 |
| meta-llama-3.1-8b-instruct | 43.1 | (-2.8, 2.3) | 628 |
|
45 |
|
46 |
+
#### Результаты на MERA
|
47 |
+
|
48 |
+
TODO
|
49 |
+
|
50 |
+
#### Результаты на llmtf_open
|
51 |
+
|
52 |
+
TODO
|
53 |
+
|
54 |
+
## How to cite:
|
55 |
|
56 |
Tikhomirov M., Chernyshev D. Facilitating large language model Russian adaptation with Learned Embedding Propagation // 2024 (will be soon)
|
57 |
|