The table will scroll to the left
Task name | Result | Metric |
---|---|---|
BPS | 0.449 | Accuracy |
LCS | 0.136 | Accuracy |
RCB | 0.333 / 0.167 | Avg. F1 / Accuracy |
USE | 0 | Grade Norm |
RWSD | 0.519 | Accuracy |
PARus | 0.498 | Accuracy |
ruTiE | 0.5 | Accuracy |
MultiQ | 0.055 / 0.014 | F1-score/EM |
ruMMLU | 0.241 | Accuracy |
CheGeKa | 0.004 / 0 | F1 / EM |
ruModAr | 0.001 | EM |
SimpleAr | 0.007 | EM |
ruMultiAr | 0.012 | EM |
MathLogicQA | 0.258 | Accuracy |
ruHumanEval | 0 / 0 / 0 | pass@k |
ruWorldTree | 0.251 / 0.225 | Avg. F1 / Accuracy |
ruOpenBookQA | 0.245 / 0.193 | Avg. F1 / Accuracy |
The table will scroll to the left
Task name | Result | Metric | ||||||||||||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
ruHHH |
0.478
|
Accuracy | ||||||||||||||||||||||||
ruHateSpeech |
0.543
|
Accuracy | ||||||||||||||||||||||||
ruDetox |
|
Overall average score (J) Assessment of the preservation of meaning (SIM) Assessment of naturalness (FL) Style Transfer Accuracy (STA) |
||||||||||||||||||||||||
ruEthics |
Table results:
[[0.071, 0.083
, 0.092, 0.075
, 0.12], |
5 MCC |
MERA
mGPT 1.3B
The mGPT architecture is based on GPT-3. We use the architecture description by Brown et al., the GPT-2 code base (Radford et al., 2019) from HuggingFace (Wolf et al., 2020) and MegatronLM (Shoeybi et al., 2020). With all the other hyperparameters equal, GPT-3 has fewer layers (Layers: 48 vs. 24) but a larger hidden size (dmodel: 1600 vs. 2048) as opposed to GPT-2. GPT-3 also alternates the classic dense and sparse attention layers (Child et al., 2019).
LM was pretrained with a total batch size of 2048 and a context window of 512 tokens. The total number of the training steps is 600k, and the models have seen 400B tokens during pretraining. The pretraining took 22 days on a cluster of 512 V100 GPUs for mGPT13B.
The pretraining corpus represents a collection of documents from Wikipedia and C4. The Wikipedia texts are extracted from the dumps (v. 20201101) with WikiExtractor (Attardi, 2015). The C4 data is downloaded using the Tensorflow datasets(Paper, 2021).
Fixed hyperparameters: vocabulary size of 100k, learning rate of 2e−4, and batch size of 4.
MIT
Code version v.1.1.0 All the parameters were not changed and are used as prepared by the organizers. Details: - 1 x NVIDIA A100 - dtype auto - Pytorch 2.1.2 + CUDA 12.1 - Transformers 4.36.2 - Context length 2048