Update README.md
Browse files
README.md
CHANGED
@@ -31,6 +31,7 @@ For text generation, like other XLNet models, a longer context will help generat
|
|
31 |
## Training data
|
32 |
|
33 |
The following is the list of data sources. Total characters is about 507M.
|
|
|
34 |
| Data | % |
|
35 |
| ------------------------------------------------- | --: |
|
36 |
| News Articles / Blogs | 58% |
|
@@ -40,6 +41,7 @@ The following is the list of data sources. Total characters is about 507M.
|
|
40 |
| Online Fiction | 1% |
|
41 |
|
42 |
The following is the distribution of different languages within the corpus.
|
|
|
43 |
| Language | % |
|
44 |
| ------------------------------------------------- | --: |
|
45 |
| Standard Chinese | 62% |
|
@@ -49,6 +51,7 @@ The following is the distribution of different languages within the corpus.
|
|
49 |
## Training procedure
|
50 |
|
51 |
Model was trained on a single TPUv3 from the official repo with the default parameters.
|
|
|
52 |
| Parameter | Value |
|
53 |
| ------------------------------------------------ | ----: |
|
54 |
| Batch Size | 32 |
|
@@ -60,6 +63,7 @@ Model was trained on a single TPUv3 from the official repo with the default para
|
|
60 |
## Eval results
|
61 |
|
62 |
Average evaluation task results over 10 runs. Comparison using the original repo model and code. Chinese models are available from [Joint Laboratory of HIT and iFLYTEK Research (HFL)](https://huggingface.co/hfl)
|
|
|
63 |
| Model | DRCD (EM/F1) | openrice-senti | lihkg-cat | wordshk-sem |
|
64 |
|:-----------:|:------------:|:--------------:|:---------:|:-----------:|
|
65 |
| Chinese | 82.8 / 91.8 | 79.8 | 70.7 | 72.0 / 78.9*|
|
|
|
31 |
## Training data
|
32 |
|
33 |
The following is the list of data sources. Total characters is about 507M.
|
34 |
+
|
35 |
| Data | % |
|
36 |
| ------------------------------------------------- | --: |
|
37 |
| News Articles / Blogs | 58% |
|
|
|
41 |
| Online Fiction | 1% |
|
42 |
|
43 |
The following is the distribution of different languages within the corpus.
|
44 |
+
|
45 |
| Language | % |
|
46 |
| ------------------------------------------------- | --: |
|
47 |
| Standard Chinese | 62% |
|
|
|
51 |
## Training procedure
|
52 |
|
53 |
Model was trained on a single TPUv3 from the official repo with the default parameters.
|
54 |
+
|
55 |
| Parameter | Value |
|
56 |
| ------------------------------------------------ | ----: |
|
57 |
| Batch Size | 32 |
|
|
|
63 |
## Eval results
|
64 |
|
65 |
Average evaluation task results over 10 runs. Comparison using the original repo model and code. Chinese models are available from [Joint Laboratory of HIT and iFLYTEK Research (HFL)](https://huggingface.co/hfl)
|
66 |
+
|
67 |
| Model | DRCD (EM/F1) | openrice-senti | lihkg-cat | wordshk-sem |
|
68 |
|:-----------:|:------------:|:--------------:|:---------:|:-----------:|
|
69 |
| Chinese | 82.8 / 91.8 | 79.8 | 70.7 | 72.0 / 78.9*|
|