Upload README.md
Browse files
README.md
CHANGED
|
@@ -10,15 +10,41 @@ license: cc-by-nc-sa-4.0
|
|
| 10 |
|
| 11 |
## Model Details
|
| 12 |
|
| 13 |
-
**Model Developers**
|
| 14 |
|
| 15 |
**Method**
|
| 16 |
-
(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
# **Model Benchmark**
|
| 19 |
|
| 20 |
-
## Open leaderboard
|
| 21 |
-
- Follow up as [link](https://huggingface.co/spaces/
|
| 22 |
|
| 23 |
| Model | Average | ARC | HellaSwag | MMLU | TruthfulQA | Ko-CommonGenV2 |
|
| 24 |
| --- | --- | --- | --- | --- | --- | --- |
|
|
@@ -27,7 +53,26 @@ license: cc-by-nc-sa-4.0
|
|
| 27 |
| beomi/Yi-Ko-6B | 48.79 | 41.04 | 53.39 | 46.28 | 41.64 | 61.63 |
|
| 28 |
| mistralai/Mistral-7B-v0.1 | 46.89 | 38.14 | 48.19 | 45.20 | 46.13 | 56.79 |
|
| 29 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 30 |
|
|
|
|
| 31 |
# Implementation Code
|
| 32 |
```python
|
| 33 |
### KO-Platypus
|
|
|
|
| 10 |
|
| 11 |
## Model Details
|
| 12 |
|
| 13 |
+
**Model Developers** Kyujin Han (kyujinpy)
|
| 14 |
|
| 15 |
**Method**
|
| 16 |
+
Using [Mergekit](https://github.com/cg123/mergekit).
|
| 17 |
+
- [upstage/SOLAR-10.7B-v1.0](https://huggingface.co/upstage/SOLAR-10.7B-v1.0)
|
| 18 |
+
- [Yhyu13/LMCocktail-10.7B-v1](Yhyu13/LMCocktail-10.7B-v1)
|
| 19 |
+
|
| 20 |
+
**Merge config**
|
| 21 |
+
```
|
| 22 |
+
slices:
|
| 23 |
+
- sources:
|
| 24 |
+
- model: upstage/SOLAR-10.7B-v1.0
|
| 25 |
+
layer_range: [0, 48]
|
| 26 |
+
- model: Yhyu13/LMCocktail-10.7B-v1
|
| 27 |
+
layer_range: [0, 48]
|
| 28 |
+
|
| 29 |
+
merge_method: slerp
|
| 30 |
+
base_model: upstage/SOLAR-10.7B-v1.0
|
| 31 |
+
|
| 32 |
+
parameters:
|
| 33 |
+
t:
|
| 34 |
+
- filter: self_attn
|
| 35 |
+
value: [0, 0.5, 0.3, 0.7, 1]
|
| 36 |
+
- filter: mlp
|
| 37 |
+
value: [1, 0.5, 0.7, 0.3, 0]
|
| 38 |
+
- value: 0.5 # fallback for rest of tensors
|
| 39 |
+
tokenizer_source: union
|
| 40 |
+
|
| 41 |
+
dtype: float16
|
| 42 |
+
```
|
| 43 |
|
| 44 |
# **Model Benchmark**
|
| 45 |
|
| 46 |
+
## Open Ko leaderboard
|
| 47 |
+
- Follow up as [Ko-link](https://huggingface.co/spaces/upstage/open-ko-llm-leaderboard).
|
| 48 |
|
| 49 |
| Model | Average | ARC | HellaSwag | MMLU | TruthfulQA | Ko-CommonGenV2 |
|
| 50 |
| --- | --- | --- | --- | --- | --- | --- |
|
|
|
|
| 53 |
| beomi/Yi-Ko-6B | 48.79 | 41.04 | 53.39 | 46.28 | 41.64 | 61.63 |
|
| 54 |
| mistralai/Mistral-7B-v0.1 | 46.89 | 38.14 | 48.19 | 45.20 | 46.13 | 56.79 |
|
| 55 |
|
| 56 |
+
- Follow up as [En-link](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).
|
| 57 |
+
|
| 58 |
+
|
| 59 |
+
## lm-evaluation-harness
|
| 60 |
+
```
|
| 61 |
+
gpt2 (pretrained=PracticeLLM/SOLAR-tail-10.7B-Merge-v1.0), limit: None, provide_description: False, num_fewshot: 0, batch_size: None
|
| 62 |
+
| Task |Version| Metric |Value | |Stderr|
|
| 63 |
+
|----------------|------:|--------|-----:|---|-----:|
|
| 64 |
+
|kobest_boolq | 0|acc |0.5021|± |0.0133|
|
| 65 |
+
| | |macro_f1|0.3343|± |0.0059|
|
| 66 |
+
|kobest_copa | 0|acc |0.6220|± |0.0153|
|
| 67 |
+
| | |macro_f1|0.6217|± |0.0154|
|
| 68 |
+
|kobest_hellaswag| 0|acc |0.4380|± |0.0222|
|
| 69 |
+
| | |acc_norm|0.5380|± |0.0223|
|
| 70 |
+
| | |macro_f1|0.4366|± |0.0222|
|
| 71 |
+
|kobest_sentineg | 0|acc |0.4962|± |0.0251|
|
| 72 |
+
| | |macro_f1|0.3316|± |0.0113|
|
| 73 |
+
```
|
| 74 |
|
| 75 |
+
|
| 76 |
# Implementation Code
|
| 77 |
```python
|
| 78 |
### KO-Platypus
|