Hugging Face's logo

qwp4w3hyb
/
Gemma-2-9B-It-SPPO-Iter3-iMat-GGUF

Text Generation
GGUF
English
gemma
SPPO
imatrix
Inference Endpoints
Model card Files Community
Gemma-2-9B-It-SPPO-Iter3-iMat-GGUF
File size: 3,919 Bytes 
e46d679
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
 
---
license: gemma
language:
- en
pipeline_tag: text-generation
tags:
- gemma
- gguf
- SPPO
- imatrix
base_model: UCLA-AGI/Gemma-2-9B-It-SPPO-Iter3
---

# Quant Infos

## Updated for all recent llama.cpp fixes (final logit soft capping+sliding window+tokenizer)

- quants done with an importance matrix for improved quantization loss
- Requantized ggufs & imatrix from hf bf16
  - initial version was based on f32 gguf provided by google, which had various issues
  - also updated for all recent llama.cpp fixes (final logit soft capping+sliding window+tokenizer)
- Wide coverage of different gguf quant types from Q\_8\_0 down to IQ1\_S
  - experimental custom quant types
    - `_L` with `--output-tensor-type f16 --token-embedding-type f16` (same as bartowski's)
- Quantized with [llama.cpp](https://github.com/ggerganov/llama.cpp) commit [5fac350b9cc49d0446fc291b9c4ad53666c77591](https://github.com/ggerganov/llama.cpp/commit/5fac350b9cc49d0446fc291b9c4ad53666c77591) (master from 2024-07-02)
- Imatrix generated with [this](https://gist.github.com/bartowski1182/eb213dccb3571f863da82e99418f81e8) multi-purpose dataset by [bartowski](https://huggingface.co/bartowski).
  ```
  ./imatrix -m $model_name-bf16.gguf -f calibration_datav3.txt -o $model_name.imatrix
  ```



---
# Original Model Card:

Self-Play Preference Optimization for Language Model Alignment (https://arxiv.org/abs/2405.00675)

# Gemma-2-9B-It-SPPO-Iter3

This model was developed using [Self-Play Preference Optimization](https://arxiv.org/abs/2405.00675) at iteration 3, based on the [google/gemma-2-9b-it](https://huggingface.co/google/gemma-2-9b-it) architecture as starting point. We utilized the prompt sets from the [openbmb/UltraFeedback](https://huggingface.co/datasets/openbmb/UltraFeedback) dataset, splited to 3 parts for 3 iterations by [snorkelai/Snorkel-Mistral-PairRM-DPO-Dataset](https://huggingface.co/datasets/snorkelai/Snorkel-Mistral-PairRM-DPO-Dataset). All responses used are synthetic.

**Terms of Use**: [Terms](https://www.kaggle.com/models/google/gemma/license/consent/verify/huggingface?returnModelRepoId=google/gemma-2-9b-it)


## Links to Other Models
- [Gemma-2-9B-It-SPPO-Iter1](https://huggingface.co/UCLA-AGI/Gemma-2-9B-It-SPPO-Iter1)
- [Gemma-2-9B-It-SPPO-Iter2](https://huggingface.co/UCLA-AGI/Gemma-2-9B-It-SPPO-Iter2)
- [Gemma-2-9B-It-SPPO-Iter3](https://huggingface.co/UCLA-AGI/Gemma-2-9B-It-SPPO-Iter3)

### Model Description

- Model type: A 8B parameter GPT-like model fine-tuned on synthetic datasets.
- Language(s) (NLP): Primarily English
- License: Apache-2.0
- Finetuned from model: google/gemma-2-9b-it


## [AlpacaEval Leaderboard Evaluation Results](https://tatsu-lab.github.io/alpaca_eval/)


|                Model                           | LC. Win Rate | Win Rate | Avg. Length |
|-------------------------------------------|:------------:|:--------:|:-----------:|
|[Gemma-2-9B-SPPO Iter1](https://huggingface.co/UCLA-AGI/Gemma-2-9B-It-SPPO-Iter1) |48.70 |40.76 | 1669
|[Gemma-2-9B-SPPO Iter2](https://huggingface.co/UCLA-AGI/Gemma-2-9B-It-SPPO-Iter2) |50.93 | 44.64 | 1759
|[Gemma-2-9B-SPPO Iter3](https://huggingface.co/UCLA-AGI/Gemma-2-9B-It-SPPO-Iter3) |**53.27** |**47.74** | 1803






### Training hyperparameters
The following hyperparameters were used during training:

- learning_rate: 5e-07
- eta: 1000
- per_device_train_batch_size: 8
- gradient_accumulation_steps: 1
- seed: 42
- distributed_type: deepspeed_zero3
- num_devices: 8
- optimizer: RMSProp 
- lr_scheduler_type: linear 
- lr_scheduler_warmup_ratio: 0.1
- num_train_epochs: 1.0



  
## Citation
```
@misc{wu2024self,
      title={Self-Play Preference Optimization for Language Model Alignment}, 
      author={Wu, Yue and Sun, Zhiqing and Yuan, Huizhuo and Ji, Kaixuan and Yang, Yiming and Gu, Quanquan},
      year={2024},
      eprint={2405.00675},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}
```