RichardErkhov
commited on
Commit
•
b38ecb4
1
Parent(s):
1028aaa
uploaded readme
Browse files
README.md
ADDED
@@ -0,0 +1,119 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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 |
+
ALMA-13B-R - GGUF
|
11 |
+
- Model creator: https://huggingface.co/haoranxu/
|
12 |
+
- Original model: https://huggingface.co/haoranxu/ALMA-13B-R/
|
13 |
+
|
14 |
+
|
15 |
+
| Name | Quant method | Size |
|
16 |
+
| ---- | ---- | ---- |
|
17 |
+
| [ALMA-13B-R.Q2_K.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q2_K.gguf) | Q2_K | 4.52GB |
|
18 |
+
| [ALMA-13B-R.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.IQ3_XS.gguf) | IQ3_XS | 4.99GB |
|
19 |
+
| [ALMA-13B-R.IQ3_S.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.IQ3_S.gguf) | IQ3_S | 5.27GB |
|
20 |
+
| [ALMA-13B-R.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q3_K_S.gguf) | Q3_K_S | 5.27GB |
|
21 |
+
| [ALMA-13B-R.IQ3_M.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.IQ3_M.gguf) | IQ3_M | 5.57GB |
|
22 |
+
| [ALMA-13B-R.Q3_K.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q3_K.gguf) | Q3_K | 5.9GB |
|
23 |
+
| [ALMA-13B-R.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q3_K_M.gguf) | Q3_K_M | 5.9GB |
|
24 |
+
| [ALMA-13B-R.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q3_K_L.gguf) | Q3_K_L | 6.45GB |
|
25 |
+
| [ALMA-13B-R.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.IQ4_XS.gguf) | IQ4_XS | 6.54GB |
|
26 |
+
| [ALMA-13B-R.Q4_0.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q4_0.gguf) | Q4_0 | 6.86GB |
|
27 |
+
| [ALMA-13B-R.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.IQ4_NL.gguf) | IQ4_NL | 6.9GB |
|
28 |
+
| [ALMA-13B-R.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q4_K_S.gguf) | Q4_K_S | 6.91GB |
|
29 |
+
| [ALMA-13B-R.Q4_K.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q4_K.gguf) | Q4_K | 7.33GB |
|
30 |
+
| [ALMA-13B-R.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q4_K_M.gguf) | Q4_K_M | 7.33GB |
|
31 |
+
| [ALMA-13B-R.Q4_1.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q4_1.gguf) | Q4_1 | 7.61GB |
|
32 |
+
| [ALMA-13B-R.Q5_0.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q5_0.gguf) | Q5_0 | 8.36GB |
|
33 |
+
| [ALMA-13B-R.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q5_K_S.gguf) | Q5_K_S | 8.36GB |
|
34 |
+
| [ALMA-13B-R.Q5_K.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q5_K.gguf) | Q5_K | 8.6GB |
|
35 |
+
| [ALMA-13B-R.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q5_K_M.gguf) | Q5_K_M | 8.6GB |
|
36 |
+
| [ALMA-13B-R.Q5_1.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q5_1.gguf) | Q5_1 | 9.1GB |
|
37 |
+
| [ALMA-13B-R.Q6_K.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q6_K.gguf) | Q6_K | 9.95GB |
|
38 |
+
| [ALMA-13B-R.Q8_0.gguf](https://huggingface.co/RichardErkhov/haoranxu_-_ALMA-13B-R-gguf/blob/main/ALMA-13B-R.Q8_0.gguf) | Q8_0 | 12.88GB |
|
39 |
+
|
40 |
+
|
41 |
+
|
42 |
+
|
43 |
+
Original model description:
|
44 |
+
---
|
45 |
+
license: mit
|
46 |
+
---
|
47 |
+
**[ALMA-R](https://arxiv.org/abs/2401.08417)** builds upon [ALMA models](https://arxiv.org/abs/2309.11674), with further LoRA fine-tuning with our proposed **Contrastive Preference Optimization (CPO)** as opposed to the Supervised Fine-tuning used in ALMA. CPO fine-tuning requires our [triplet preference data](https://huggingface.co/datasets/haoranxu/ALMA-R-Preference) for preference learning. ALMA-R now can matches or even exceeds GPT-4 or WMT winners!
|
48 |
+
```
|
49 |
+
@misc{xu2024contrastive,
|
50 |
+
title={Contrastive Preference Optimization: Pushing the Boundaries of LLM Performance in Machine Translation},
|
51 |
+
author={Haoran Xu and Amr Sharaf and Yunmo Chen and Weiting Tan and Lingfeng Shen and Benjamin Van Durme and Kenton Murray and Young Jin Kim},
|
52 |
+
year={2024},
|
53 |
+
eprint={2401.08417},
|
54 |
+
archivePrefix={arXiv},
|
55 |
+
primaryClass={cs.CL}
|
56 |
+
}
|
57 |
+
```
|
58 |
+
```
|
59 |
+
@misc{xu2023paradigm,
|
60 |
+
title={A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models},
|
61 |
+
author={Haoran Xu and Young Jin Kim and Amr Sharaf and Hany Hassan Awadalla},
|
62 |
+
year={2023},
|
63 |
+
eprint={2309.11674},
|
64 |
+
archivePrefix={arXiv},
|
65 |
+
primaryClass={cs.CL}
|
66 |
+
}
|
67 |
+
```
|
68 |
+
# Download ALMA(-R) Models and Dataset 🚀
|
69 |
+
|
70 |
+
We release six translation models presented in the paper:
|
71 |
+
- ALMA-7B
|
72 |
+
- ALMA-7B-LoRA
|
73 |
+
- **ALMA-7B-R (NEW!)**: Further LoRA fine-tuning upon ALMA-7B-LoRA with contrastive preference optimization.
|
74 |
+
- ALMA-13B
|
75 |
+
- ALMA-13B-LoRA
|
76 |
+
- **ALMA-13B-R (NEW!)**: Further LoRA fine-tuning upon ALMA-13B-LoRA with contrastive preference optimization (BEST MODEL!).
|
77 |
+
|
78 |
+
Model checkpoints are released at huggingface:
|
79 |
+
| Models | Base Model Link | LoRA Link |
|
80 |
+
|:-------------:|:---------------:|:---------:|
|
81 |
+
| ALMA-7B | [haoranxu/ALMA-7B](https://huggingface.co/haoranxu/ALMA-7B) | - |
|
82 |
+
| ALMA-7B-LoRA | [haoranxu/ALMA-7B-Pretrain](https://huggingface.co/haoranxu/ALMA-7B-Pretrain) | [haoranxu/ALMA-7B-Pretrain-LoRA](https://huggingface.co/haoranxu/ALMA-7B-Pretrain-LoRA) |
|
83 |
+
| **ALMA-7B-R (NEW!)** | [haoranxu/ALMA-7B-R (LoRA merged)](https://huggingface.co/haoranxu/ALMA-7B-R) | - |
|
84 |
+
| ALMA-13B | [haoranxu/ALMA-13B](https://huggingface.co/haoranxu/ALMA-13B) | - |
|
85 |
+
| ALMA-13B-LoRA | [haoranxu/ALMA-13B-Pretrain](https://huggingface.co/haoranxu/ALMA-13B-Pretrain) | [haoranxu/ALMA-13B-Pretrain-LoRA](https://huggingface.co/haoranxu/ALMA-13B-Pretrain-LoRA) |
|
86 |
+
| **ALMA-13B-R (NEW!)** | [haoranxu/ALMA-13B-R (LoRA merged)](https://huggingface.co/haoranxu/ALMA-13B-R) | - |
|
87 |
+
|
88 |
+
**Note that `ALMA-7B-Pretrain` and `ALMA-13B-Pretrain` are NOT translation models. They only experience stage 1 monolingual fine-tuning (20B tokens for the 7B model and 12B tokens for the 13B model), and should be utilized in conjunction with their LoRA models.**
|
89 |
+
|
90 |
+
Datasets used by ALMA and ALMA-R are also released at huggingface now (NEW!)
|
91 |
+
| Datasets | Train / Validation| Test |
|
92 |
+
|:-------------:|:---------------:|:---------:|
|
93 |
+
| Human-Written Parallel Data (ALMA) | [train and validation](https://huggingface.co/datasets/haoranxu/ALMA-Human-Parallel) | [WMT'22](https://huggingface.co/datasets/haoranxu/WMT22-Test) |
|
94 |
+
| Triplet Preference Data | [train](https://huggingface.co/datasets/haoranxu/ALMA-R-Preference) | [WMT'22](https://huggingface.co/datasets/haoranxu/WMT22-Test) and [WMT'23](https://huggingface.co/datasets/haoranxu/WMT23-Test) |
|
95 |
+
|
96 |
+
|
97 |
+
A quick start to use our best system (ALMA-13B-R) for translation. An example of translating "我爱机器翻译。" into English:
|
98 |
+
```
|
99 |
+
import torch
|
100 |
+
from transformers import AutoModelForCausalLM
|
101 |
+
from transformers import AutoTokenizer
|
102 |
+
|
103 |
+
# Load base model and LoRA weights
|
104 |
+
model = AutoModelForCausalLM.from_pretrained("haoranxu/ALMA-13B-R", torch_dtype=torch.float16, device_map="auto")
|
105 |
+
tokenizer = AutoTokenizer.from_pretrained("haoranxu/ALMA-13B-R", padding_side='left')
|
106 |
+
|
107 |
+
# Add the source sentence into the prompt template
|
108 |
+
prompt="Translate this from Chinese to English:\nChinese: 我爱机器翻译。\nEnglish:"
|
109 |
+
input_ids = tokenizer(prompt, return_tensors="pt", padding=True, max_length=40, truncation=True).input_ids.cuda()
|
110 |
+
|
111 |
+
# Translation
|
112 |
+
with torch.no_grad():
|
113 |
+
generated_ids = model.generate(input_ids=input_ids, num_beams=5, max_new_tokens=20, do_sample=True, temperature=0.6, top_p=0.9)
|
114 |
+
outputs = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
|
115 |
+
print(outputs)
|
116 |
+
```
|
117 |
+
|
118 |
+
Please find more details in our [GitHub repository](https://github.com/fe1ixxu/ALMA)
|
119 |
+
|