Upload ChineseBabyLM best model artifacts
Browse files- README.md +48 -0
- config.json +28 -0
- model.safetensors +3 -0
- tokenizer.json +0 -0
- tokenizer_config.json +10 -0
README.md
ADDED
|
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- zh
|
| 4 |
+
license: other
|
| 5 |
+
library_name: transformers
|
| 6 |
+
pipeline_tag: fill-mask
|
| 7 |
+
tags:
|
| 8 |
+
- chinesebabylm
|
| 9 |
+
- bert
|
| 10 |
+
- masked-language-modeling
|
| 11 |
+
- baby-lm
|
| 12 |
+
- control
|
| 13 |
+
- adamw
|
| 14 |
+
---
|
| 15 |
+
|
| 16 |
+
# ChineseBabyLM Control AdamW BERT MLM 9.53M
|
| 17 |
+
|
| 18 |
+
Small from-scratch BERT-style masked language model trained as the Round 20 DeepScientist control-arm AdamW no-bias-correction pilot.
|
| 19 |
+
|
| 20 |
+
## Model Details
|
| 21 |
+
|
| 22 |
+
- Architecture: `BertForMaskedLM`
|
| 23 |
+
- Model type: masked language model / encoder-only BERT
|
| 24 |
+
- Parameters: 9,527,714
|
| 25 |
+
- Layers: 4
|
| 26 |
+
- Attention heads: 6
|
| 27 |
+
- Hidden size: 384
|
| 28 |
+
- Vocabulary size: 5,666
|
| 29 |
+
- Max position embeddings: 256
|
| 30 |
+
|
| 31 |
+
## Training Data
|
| 32 |
+
|
| 33 |
+
Chinese BabyLM official 10k stratified sample; control-arm Round 20 AdamW no-bias-correction pilot.
|
| 34 |
+
|
| 35 |
+
## Evaluation
|
| 36 |
+
|
| 37 |
+
Official zero-shot evaluator results:
|
| 38 |
+
|
| 39 |
+
| Task | Accuracy |
|
| 40 |
+
|---|---:|
|
| 41 |
+
| ZhoBLiMP | 59.74 |
|
| 42 |
+
| Hanzi structure | 47.00 |
|
| 43 |
+
| Hanzi pinyin | 31.00 |
|
| 44 |
+
| Mean available tasks | 45.91 |
|
| 45 |
+
|
| 46 |
+
## Package Source
|
| 47 |
+
|
| 48 |
+
This model was packaged from `control_round20_adamw_no_bias_correction_bert_mlm_9p53m` in the local ChineseBabyLM best-model archive.
|
config.json
ADDED
|
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_cross_attention": false,
|
| 3 |
+
"architectures": [
|
| 4 |
+
"BertForMaskedLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_probs_dropout_prob": 0.1,
|
| 7 |
+
"bos_token_id": null,
|
| 8 |
+
"classifier_dropout": null,
|
| 9 |
+
"dtype": "float32",
|
| 10 |
+
"eos_token_id": null,
|
| 11 |
+
"hidden_act": "gelu",
|
| 12 |
+
"hidden_dropout_prob": 0.1,
|
| 13 |
+
"hidden_size": 384,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 1536,
|
| 16 |
+
"is_decoder": false,
|
| 17 |
+
"layer_norm_eps": 1e-12,
|
| 18 |
+
"max_position_embeddings": 256,
|
| 19 |
+
"model_type": "bert",
|
| 20 |
+
"num_attention_heads": 6,
|
| 21 |
+
"num_hidden_layers": 4,
|
| 22 |
+
"pad_token_id": 0,
|
| 23 |
+
"tie_word_embeddings": true,
|
| 24 |
+
"transformers_version": "5.3.0.dev0",
|
| 25 |
+
"type_vocab_size": 2,
|
| 26 |
+
"use_cache": true,
|
| 27 |
+
"vocab_size": 5666
|
| 28 |
+
}
|
model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:df7b94124948eb0cebdfe4cdcd49f7c47748d7363ab533e83635fcdf021ff3cf
|
| 3 |
+
size 38119320
|
tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
tokenizer_config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"backend": "tokenizers",
|
| 3 |
+
"cls_token": "[CLS]",
|
| 4 |
+
"mask_token": "[MASK]",
|
| 5 |
+
"model_max_length": 128,
|
| 6 |
+
"pad_token": "[PAD]",
|
| 7 |
+
"sep_token": "[SEP]",
|
| 8 |
+
"tokenizer_class": "TokenizersBackend",
|
| 9 |
+
"unk_token": "[UNK]"
|
| 10 |
+
}
|