Downloads · 30 days
210
6% of all-time downloads
vmajor_-_Orca2-13B-selfmerge-26B-gguf
RichardErkhov/vmajor_-_Orca2-13B-selfmerge-26B-gguf
vmajor_-_Orca2-13B-selfmerge-26B-gguf is a machine learning model from RichardErkhov. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
All-time downloads
3.4K
Public
Repo size
148 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.gguf171 GB · 100%
At a glance
- Access
- Public
- Created
- Aug 4, 2024
- Updated
- Aug 5, 2024
- SHA
- a8fbb34d
Model card
From the Hugging Face model README
Quantization made by Richard Erkhov.
Orca2-13B-selfmerge-26B - GGUF
- Model creator: https://huggingface.co/vmajor/
- Original model: https://huggingface.co/vmajor/Orca2-13B-selfmerge-26B/
Original model description:
license: ms-pl tags:
- merge
This model is a result of merging Orca2-13B with itself using 'mergekit-legacy'. Merge parameters were --weight 0.5 --density 0.5
This merged model showed marginal improvement in perplexity scores:
| Model | Perplexity |
|---|---|
| microsoft/Orca-2-13b | 7.595028877258301 |
| vmajor/Orca2-13B-selfmerge-26B | 7.550178050994873 |
| vmajor/Orca2-13B-selfmerge-39B | NC |
Benchmark Results
The following table summarizes the model performance across a range of benchmarks:
| Model | Average | ARC | HellaSwag | MMLU | TruthfulQA | Winogrande | GSM8K |
|---|---|---|---|---|---|---|---|
| microsoft/Orca-2-13b | 58.64 | 60.67 | 79.81 | 60.37 | 56.41 | 76.64 | 17.97 |
| vmajor/Orca2-13B-selfmerge-26B | 62.24 | 60.84 | 79.84 | 60.32 | 56.38 | 76.87 | 39.2 |
| vmajor/Orca2-13B-selfmerge-39B | 62.24 | 60.84 | 79.84 | 60.32 | 56.38 | 76.87 | 39.2 |
Interestingly the GSM8K performance more than doubled with the first self merge. Second self merge resulting in the 39B model did not produce any further gains.
license: ms-pl
- Created
- Aug 4, 2024
- Updated
- Aug 5, 2024