RichardErkhov
commited on
Commit
•
05156fe
1
Parent(s):
3a05123
uploaded readme
Browse files
README.md
ADDED
@@ -0,0 +1,103 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
Quantization made by Richard Erkhov.
|
2 |
+
|
3 |
+
[Github](https://github.com/RichardErkhov)
|
4 |
+
|
5 |
+
[Discord](https://discord.gg/pvy7H8DZMG)
|
6 |
+
|
7 |
+
[Request more models](https://github.com/RichardErkhov/quant_request)
|
8 |
+
|
9 |
+
|
10 |
+
zephyr-7b-sft-full-SPIN-iter3 - GGUF
|
11 |
+
- Model creator: https://huggingface.co/UCLA-AGI/
|
12 |
+
- Original model: https://huggingface.co/UCLA-AGI/zephyr-7b-sft-full-SPIN-iter3/
|
13 |
+
|
14 |
+
|
15 |
+
| Name | Quant method | Size |
|
16 |
+
| ---- | ---- | ---- |
|
17 |
+
| [zephyr-7b-sft-full-SPIN-iter3.Q2_K.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.Q2_K.gguf) | Q2_K | 2.53GB |
|
18 |
+
| [zephyr-7b-sft-full-SPIN-iter3.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.IQ3_XS.gguf) | IQ3_XS | 2.81GB |
|
19 |
+
| [zephyr-7b-sft-full-SPIN-iter3.IQ3_S.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.IQ3_S.gguf) | IQ3_S | 2.96GB |
|
20 |
+
| [zephyr-7b-sft-full-SPIN-iter3.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.Q3_K_S.gguf) | Q3_K_S | 2.95GB |
|
21 |
+
| [zephyr-7b-sft-full-SPIN-iter3.IQ3_M.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.IQ3_M.gguf) | IQ3_M | 3.06GB |
|
22 |
+
| [zephyr-7b-sft-full-SPIN-iter3.Q3_K.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.Q3_K.gguf) | Q3_K | 3.28GB |
|
23 |
+
| [zephyr-7b-sft-full-SPIN-iter3.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.Q3_K_M.gguf) | Q3_K_M | 3.28GB |
|
24 |
+
| [zephyr-7b-sft-full-SPIN-iter3.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.Q3_K_L.gguf) | Q3_K_L | 3.56GB |
|
25 |
+
| [zephyr-7b-sft-full-SPIN-iter3.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.IQ4_XS.gguf) | IQ4_XS | 3.67GB |
|
26 |
+
| [zephyr-7b-sft-full-SPIN-iter3.Q4_0.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.Q4_0.gguf) | Q4_0 | 3.83GB |
|
27 |
+
| [zephyr-7b-sft-full-SPIN-iter3.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.IQ4_NL.gguf) | IQ4_NL | 3.87GB |
|
28 |
+
| [zephyr-7b-sft-full-SPIN-iter3.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.Q4_K_S.gguf) | Q4_K_S | 3.86GB |
|
29 |
+
| [zephyr-7b-sft-full-SPIN-iter3.Q4_K.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.Q4_K.gguf) | Q4_K | 4.07GB |
|
30 |
+
| [zephyr-7b-sft-full-SPIN-iter3.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.Q4_K_M.gguf) | Q4_K_M | 4.07GB |
|
31 |
+
| [zephyr-7b-sft-full-SPIN-iter3.Q4_1.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.Q4_1.gguf) | Q4_1 | 4.24GB |
|
32 |
+
| [zephyr-7b-sft-full-SPIN-iter3.Q5_0.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.Q5_0.gguf) | Q5_0 | 4.65GB |
|
33 |
+
| [zephyr-7b-sft-full-SPIN-iter3.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.Q5_K_S.gguf) | Q5_K_S | 4.65GB |
|
34 |
+
| [zephyr-7b-sft-full-SPIN-iter3.Q5_K.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.Q5_K.gguf) | Q5_K | 4.78GB |
|
35 |
+
| [zephyr-7b-sft-full-SPIN-iter3.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.Q5_K_M.gguf) | Q5_K_M | 4.78GB |
|
36 |
+
| [zephyr-7b-sft-full-SPIN-iter3.Q5_1.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.Q5_1.gguf) | Q5_1 | 5.07GB |
|
37 |
+
| [zephyr-7b-sft-full-SPIN-iter3.Q6_K.gguf](https://huggingface.co/RichardErkhov/UCLA-AGI_-_zephyr-7b-sft-full-SPIN-iter3-gguf/blob/main/zephyr-7b-sft-full-SPIN-iter3.Q6_K.gguf) | Q6_K | 5.53GB |
|
38 |
+
|
39 |
+
|
40 |
+
|
41 |
+
|
42 |
+
Original model description:
|
43 |
+
---
|
44 |
+
license: mit
|
45 |
+
datasets:
|
46 |
+
- UCLA-AGI/SPIN_iter3
|
47 |
+
language:
|
48 |
+
- en
|
49 |
+
pipeline_tag: text-generation
|
50 |
+
---
|
51 |
+
Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models (https://arxiv.org/abs/2401.01335)
|
52 |
+
|
53 |
+
# zephyr-7b-sft-full-spin-iter3
|
54 |
+
|
55 |
+
This model is a self-play fine-tuned model at iteration 3 from [alignment-handbook/zephyr-7b-sft-full](https://huggingface.co/alignment-handbook/zephyr-7b-sft-full) using synthetic data based on on the [HuggingFaceH4/ultrachat_200k](https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k) dataset.
|
56 |
+
|
57 |
+
## Model Details
|
58 |
+
|
59 |
+
### Model Description
|
60 |
+
|
61 |
+
- Model type: A 7B parameter GPT-like model fine-tuned on synthetic datasets.
|
62 |
+
- Language(s) (NLP): Primarily English
|
63 |
+
- License: MIT
|
64 |
+
- Finetuned from model: alignment-handbook/zephyr-7b-sft-full (based on mistralai/Mistral-7B-v0.1)
|
65 |
+
|
66 |
+
### Training hyperparameters
|
67 |
+
The following hyperparameters were used during training:
|
68 |
+
|
69 |
+
- learning_rate: 1e-07
|
70 |
+
- train_batch_size: 8
|
71 |
+
- seed: 42
|
72 |
+
- distributed_type: multi-GPU
|
73 |
+
- num_devices: 8
|
74 |
+
- total_train_batch_size: 64
|
75 |
+
- optimizer: RMSProp
|
76 |
+
- lr_scheduler_type: linear
|
77 |
+
- lr_scheduler_warmup_ratio: 0.1
|
78 |
+
- num_epochs: 2.0
|
79 |
+
|
80 |
+
## [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
|
81 |
+
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_UCLA-AGI__test_final)
|
82 |
+
| Metric | Value |
|
83 |
+
|-----------------------|---------------------------|
|
84 |
+
| Avg. | 63.70 |
|
85 |
+
| ARC (25-shot) | 66.13 |
|
86 |
+
| HellaSwag (10-shot) | 85.85 |
|
87 |
+
| MMLU (5-shot) | 61.51 |
|
88 |
+
| TruthfulQA (0-shot) | 57.89 |
|
89 |
+
| Winogrande (5-shot) | 76.64 |
|
90 |
+
| GSM8K (5-shot) | 34.19 |
|
91 |
+
|
92 |
+
## Citation
|
93 |
+
```
|
94 |
+
@misc{chen2024selfplay,
|
95 |
+
title={Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models},
|
96 |
+
author={Zixiang Chen and Yihe Deng and Huizhuo Yuan and Kaixuan Ji and Quanquan Gu},
|
97 |
+
year={2024},
|
98 |
+
eprint={2401.01335},
|
99 |
+
archivePrefix={arXiv},
|
100 |
+
primaryClass={cs.LG}
|
101 |
+
}
|
102 |
+
```
|
103 |
+
|