Add results
Browse files
README.md
CHANGED
|
@@ -1,3 +1,60 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
+
datasets:
|
| 4 |
+
- wenbopan/Fusang-v1
|
| 5 |
+
- wenbopan/OpenOrca-zh-20k
|
| 6 |
+
language:
|
| 7 |
+
- zh
|
| 8 |
+
- en
|
| 9 |
---
|
| 10 |
+
|
| 11 |
+
# Fi-9B
|
| 12 |
+
|
| 13 |
+
Fi-9B is an improved [Yi-9B-200K](https://huggingface.co/01-ai/Yi-9B-200K) with extensive instruction tuning on [Fusang-V1](https://huggingface.co/datasets/wenbopan/Fusang-v1).
|
| 14 |
+
|
| 15 |
+
Compare to Yi-9B-200K, Fi-9B gains greater capability at various downstream tasks and long-context modeling thanks to large-scale synthestic data in Fusang-V1.
|
| 16 |
+
|
| 17 |
+
## Performance
|
| 18 |
+
|
| 19 |
+
### Fact-based Evaluation
|
| 20 |
+
|
| 21 |
+
Fi is competitive amongst all models at ~9B size range:
|
| 22 |
+
|
| 23 |
+
| **Metric** | **winogrande** | **hellaswag** | **truthfulqa** | **ai2_arc** |
|
| 24 |
+
| ---------- | -------------- | ------------- | -------------- | ----------- |
|
| 25 |
+
| Yi-9B-200K | 0.7167 | 0.5672 | 0.3380 | 0.6925 |
|
| 26 |
+
| Fi-9B-200K | 0.7111 | **0.5728** | **0.4086** | **0.7258** |
|
| 27 |
+
|
| 28 |
+
### Long-context Modeling
|
| 29 |
+
|
| 30 |
+
Fi make even further progresses than Yi-9B-200K, which is already impressive in terms of handling long-range tasks:
|
| 31 |
+
|
| 32 |
+
Results on LongBench:
|
| 33 |
+
|
| 34 |
+
| **Name** | **Average_zh** | **Average_en** | **Code Completion** |
|
| 35 |
+
|-----------------|----------------|----------------|---------------------|
|
| 36 |
+
| Yi-9B-200K | 30.288 | 36.7071 | 72.2 |
|
| 37 |
+
| Fi-9B-200K | **41.092** | **40.9536** | 46.0 |
|
| 38 |
+
|
| 39 |
+
Detailed score decomposition on LongBench
|
| 40 |
+
|
| 41 |
+
| **Name** | **Few-shot Learning_en** | **Synthetic Tasks_en** | **Single-Doc QA_en** | **Multi-Doc QA_en** | **Summarization_en** | **Few-shot Learning_zh** | **Synthetic Tasks_zh** | **Single-Doc QA_zh** | **Multi-Doc QA_zh** | **Summarization_zh** |
|
| 42 |
+
|-----------------|--------------------------|------------------------|----------------------|---------------------|----------------------|--------------------------|------------------------|----------------------|---------------------|----------------------|
|
| 43 |
+
| Yi-9B-200K | 60.6 | 22.8 | 30.9 | 38.9 | 25.8 | 46.5 | 28.0 | 49.6 | 17.7 | 9.7 |
|
| 44 |
+
| Fi-9B-200K | **63.8** | **40.2** | **36.2** | 38.0 | **26.3** | 30.0 | **75.1** | **55.6** | **30.7** | **14.1** |
|
| 45 |
+
|
| 46 |
+
<!--### Performance on Preference TODO-->
|
| 47 |
+
|
| 48 |
+
### Bilingual Ability
|
| 49 |
+
|
| 50 |
+
Coming soon...
|
| 51 |
+
|
| 52 |
+
## How to use Fi
|
| 53 |
+
|
| 54 |
+
Coming soon...
|
| 55 |
+
|
| 56 |
+
## Current Limitations
|
| 57 |
+
|
| 58 |
+
This version of Fi-9B may not be able to stop generation in some scenarios. I will fix that soon.
|
| 59 |
+
|
| 60 |
+
Compare to the original Yi-9B-200K, Fi-9B has degraded ability for code completion. This may due to lack of raw code data during instruction tuning.
|