Commit
•
a6e5daf
1
Parent(s):
a88b991
upload files
Browse files- MODEL_LICENSE.pdf +0 -0
- README.md +152 -0
- config.json +27 -0
- configuration_xverse.py +121 -0
- modeling_xverse.py +767 -0
- pytorch_model-00001-of-00003.bin +3 -0
- pytorch_model-00002-of-00003.bin +3 -0
- pytorch_model-00003-of-00003.bin +3 -0
- pytorch_model.bin.index.json +410 -0
- special_tokens_map.json +23 -0
- tokenizer.json +0 -0
- tokenizer_config.json +5 -0
MODEL_LICENSE.pdf
ADDED
|
Binary file (315 kB). View file
|
|
|
README.md
ADDED
|
@@ -0,0 +1,152 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
|
| 4 |
+
inference: false
|
| 5 |
+
|
| 6 |
+
---
|
| 7 |
+
|
| 8 |
+
# XVERSE-13B
|
| 9 |
+
|
| 10 |
+
## 模型介绍
|
| 11 |
+
|
| 12 |
+
**XVERSE-13B** 是由深圳元象科技自主研发的支持多语言的大语言模型（Large Language Model），主要特点如下：
|
| 13 |
+
|
| 14 |
+
- **模型结构**：XVERSE-13B 使用主流 Decoder-only 的标准 Transformer 网络结构，支持 8K 的上下文长度（Context Length），为同尺寸模型中最长，能满足更长的多轮对话、知识问答与摘要等需求，模型应用场景更广泛。
|
| 15 |
+
- **训练数据**：构建了 1.4 万亿 token 的高质量、多样化的数据对模型进行充分训练，包含中、英、俄、西等 40 多种语言，通过精细化设置不同类型数据的采样比例，使得中英两种语言表现优异，也能兼顾其他语言效果。
|
| 16 |
+
- **分词**：基于 BPE（Byte-Pair Encoding）算法，使用上百 GB 语料训练了一个词表大小为 100,278 的分词器，能够同时支持多语言，而无需额外扩展词表。
|
| 17 |
+
- **训练框架**：经过对算子、通信、并行策略及调度等的优化，千卡集群上的峰值算力利用率达到 58.5%，达到业界前列。
|
| 18 |
+
|
| 19 |
+
## Model Introduction
|
| 20 |
+
|
| 21 |
+
**XVERSE-13B** is a multilingual large language model, independently developed by Shenzhen Yuanxiang Technology. Its key features are as follows:
|
| 22 |
+
|
| 23 |
+
- **Model Structure**: XVERSE-13B uses the mainstream Decoder-only Transformer network structure, supports 8k context length, the longest one among models of the same size, which can meet the need of longer multi-round dialogues, knowledge question-answering, and summarization. This makes the model more versatile in application scenarios.
|
| 24 |
+
- **Training Data**: The model has been thoroughly trained on a diversified and high-quality dataset consisting of 1.4 trillion of tokens, including more than 40 languages such as Chinese, English, Russian, and Spanish. The sampling ratio of different types of data is finely set, which makes the performance of Chinese and English excellent, and also takes into account the effect of other languages.
|
| 25 |
+
- **Tokenization**: Based on the BPE (Byte-Pair Encoding) algorithm, a tokenizer with a vocabulary size of 100,278 has been trained using hundreds of gigabytes of language data. This tokenizer is capable of supporting multilingual without the need for additional vocabulary expansion.
|
| 26 |
+
- **Training Framework**: Through optimizations in operators, communication, parallel strategies, and scheduling, the peak computing power utilization rate on a thousand-card cluster reaches 58.5%, which is one of the highest in the industry.
|
| 27 |
+
|
| 28 |
+
## 评测结果
|
| 29 |
+
|
| 30 |
+
为验证模型的各项能力，我们选取了多个学科综合能力评测集，包括 [MMLU](https://arxiv.org/abs/2009.03300)（英文）、 [C-Eval](https://cevalbenchmark.com/)（中文）、[AGIEval](https://arxiv.org/abs/2304.06364)（中英） 、[GAOKAO-Bench](https://github.com/OpenLMLab/GAOKAO-Bench)（中英）、[GAOKAO-English](https://github.com/ExpressAI/AI-Gaokao)（英文），评测结果如下：
|
| 31 |
+
|
| 32 |
+
| 模型\数据集 | MMLU | C-Eval | AGIEval<sup>1</sup> | GAOKAO-Bench<sup>1</sup> | GAOKAO-English<sup>1</sup> |
|
| 33 |
+
| :------------------------: | :--------------: | :--------------: | :-----------------: | :----------------------: | :------------------------: |
|
| 34 |
+
| Baichuan-13B | 51.6<sup>2</sup> | 53.6<sup>3</sup> | 40.5 | 45.9 | 56.9 |
|
| 35 |
+
| Llama-1-13B | 46.9<sup>4</sup> | 28.8 | 27.3 | 26.4 | 38.1 |
|
| 36 |
+
| Llama-2-13B | 54.8<sup>4</sup> | 35.6 | 33.4 | 35.4 | 60.6 |
|
| 37 |
+
| moss-moon-003-base (16B) | 24.7 | 33.1<sup>3</sup> | 26.8 | 28.5 | 34.7 |
|
| 38 |
+
| OpenLLaMA-13B | 42.4 | 24.7 | 24.0 | 25.6 | 33.3 |
|
| 39 |
+
| OPT-13B | 25.2 | 25.0 | 24.2 | 24.4 | 31.1 |
|
| 40 |
+
| Pythia-12B | 25.1 | 26.2 | 25.3 | 25.3 | 26.8 |
|
| 41 |
+
| Ziya-LLaMA-13B-Pretrain-v1 | 43.9 | 30.2 | 27.2 | 26.4 | 37.6 |
|
| 42 |
+
| **XVERSE-13B** | **55.1** | **54.7** | **41.4** | **53.9** | **66.5** |
|
| 43 |
+
|
| 44 |
+
> <sup>1：只针对其中的单项选择题进行测试，即排除了填空题、开放性问题和多项选择题</sup>
|
| 45 |
+
> <sup>2：来源于 [Baichuan-13B](https://github.com/baichuan-inc/Baichuan-13B) 的汇报结果</sup>
|
| 46 |
+
> <sup>3：来源于 [C-Eval](https://cevalbenchmark.com/) 的汇报结果</sup>
|
| 47 |
+
> <sup>4：来源于[Llama 2 论文](https://arxiv.org/abs/2307.09288)的汇报结果</sup>
|
| 48 |
+
>
|
| 49 |
+
> 对于 MMLU ，我们采用作者提供的[评测工具](https://github.com/hendrycks/test)，C-Eval、AGIEval、GAOKAO-Bench、GAOKAO-English 与 MMLU 的评测方式相同，且统一采用 **5-shot** 构造测试样本。
|
| 50 |
+
|
| 51 |
+
|
| 52 |
+
## Model Evaluation
|
| 53 |
+
|
| 54 |
+
In order to validate the various abilities of the model, we have chosen several comprehensive capability benchmarks across multiple disciplines, including [MMLU](https://arxiv.org/abs/2009.03300) (English), [C-Eval](https://cevalbenchmark.com/) (Chinese), [AGIEval](https://arxiv.org/abs/2304.06364) (Chinese and English), [GAOKAO-Bench](https://github.com/OpenLMLab/GAOKAO-Bench) (Chinese and English), [GAOKAO-English](https://github.com/ExpressAI/AI-Gaokao) (English), the evaluation results are as follows:
|
| 55 |
+
|
| 56 |
+
| Models\Datasets | MMLU | C-Eval | AGIEval<sup>1</sup> | GAOKAO-Bench<sup>1</sup> | GAOKAO-English<sup>1</sup> |
|
| 57 |
+
| :------------------------: | :--------------: | :--------------: | :-----------------: | :----------------------: | :------------------------: |
|
| 58 |
+
| Baichuan-13B | 51.6<sup>2</sup> | 53.6<sup>3</sup> | 40.5 | 45.9 | 56.9 |
|
| 59 |
+
| Llama-1-13B | 46.9<sup>4</sup> | 28.8 | 27.3 | 26.4 | 38.1 |
|
| 60 |
+
| Llama-2-13B | 54.8<sup>4</sup> | 35.6 | 33.4 | 35.4 | 60.6 |
|
| 61 |
+
| moss-moon-003-base (16B) | 24.7 | 33.1<sup>3</sup> | 26.8 | 28.5 | 34.7 |
|
| 62 |
+
| OpenLLaMA-13B | 42.4 | 24.7 | 24.0 | 25.6 | 33.3 |
|
| 63 |
+
| OPT-13B | 25.2 | 25.0 | 24.2 | 24.4 | 31.1 |
|
| 64 |
+
| Pythia-12B | 25.1 | 26.2 | 25.3 | 25.3 | 26.8 |
|
| 65 |
+
| Ziya-LLaMA-13B-Pretrain-v1 | 43.9 | 30.2 | 27.2 | 26.4 | 37.6 |
|
| 66 |
+
| **XVERSE-13B** | **55.1** | **54.7** | **41.4** | **53.9** | **66.5** |
|
| 67 |
+
|
| 68 |
+
> <sup>1: Tests are conducted only on single-answer multiple-choice questions, thus excluding fill-in-the-blanks, open-ended questions, and multiple-answer multiple-choice questions.</sup>
|
| 69 |
+
> <sup>2: Reporting results from [Baichuan-13B](https://github.com/baichuan-inc/Baichuan-13B).</sup>
|
| 70 |
+
> <sup>3: Reporting results from [C-Eval](https://cevalbenchmark.com/).</sup>
|
| 71 |
+
> <sup>4: Reporting results from [Llama 2](https://arxiv.org/abs/2307.09288).</sup>
|
| 72 |
+
>
|
| 73 |
+
> For MMLU, we adopt the [evaluation tools](https://github.com/hendrycks/test) provided by the authors, C-Eval, AGIEval, GAOKAO-Bench, GAOKAO-English are the same as MMLU, and uniformly use **5-shot** to construct the test samples.
|
| 74 |
+
|
| 75 |
+
### MMLU 各类别指标
|
| 76 |
+
|
| 77 |
+
MMLU Category Results
|
| 78 |
+
|
| 79 |
+
| 模型\类别 | Average | STEM | Social Science | Humanities | Others |
|
| 80 |
+
| :------------------------: | :------: | :------: | :------------: | :--------: | :------: |
|
| 81 |
+
| Baichuan-13B | 51.6 | 41.6 | 60.9 | 47.4 | 58.5 |
|
| 82 |
+
| Llama-1-13B | 46.9 | 35.8 | 53.8 | 45.0 | 53.3 |
|
| 83 |
+
| Llama-2-13B | 54.8 | 44.1 | 62.6 | 52.8 | 61.1 |
|
| 84 |
+
| moss-moon-003-base (16B) | 24.7 | 23.0 | 24.0 | 25.2 | 26.3 |
|
| 85 |
+
| OpenLLaMA-13B | 42.4 | 34.7 | 48.6 | 40.0 | 47.1 |
|
| 86 |
+
| OPT-13B | 25.2 | 23.9 | 24.1 | 25.9 | 26.3 |
|
| 87 |
+
| Pythia-12B | 25.1 | 24.8 | 23.0 | 26.1 | 26.0 |
|
| 88 |
+
| Ziya-LLaMA-13B-Pretrain-v1 | 43.9 | 36.3 | 48.8 | 41.1 | 50.3 |
|
| 89 |
+
| **XVERSE-13B** | **55.1** | **44.5** | **64.4** | **50.5** | **62.9** |
|
| 90 |
+
|
| 91 |
+
### C-Eval 各类别指标
|
| 92 |
+
|
| 93 |
+
C-Eval Category Results
|
| 94 |
+
|
| 95 |
+
| 模型\类别 | Average | STEM | Social Science | Humanities | Others |
|
| 96 |
+
| :------------------------: | :------: | :------: | :------------: | :--------: | :------: |
|
| 97 |
+
| Baichuan-13B | 53.6 | 47.0 | 66.8 | 57.3 | 49.8 |
|
| 98 |
+
| Llama-1-13B | 28.8 | 27.5 | 33.9 | 27.7 | 27.7 |
|
| 99 |
+
| Llama-2-13B | 35.6 | 34.5 | 39.8 | 36.2 | 33.2 |
|
| 100 |
+
| moss-moon-003-base (16B) | 33.1 | 31.6 | 37.0 | 33.4 | 32.1 |
|
| 101 |
+
| OpenLLaMA-13B | 24.7 | 25.5 | 23.5 | 24.2 | 24.7 |
|
| 102 |
+
| OPT-13B | 25.0 | 24.4 | 24.6 | 25.9 | 25.4 |
|
| 103 |
+
| Pythia-12B | 26.2 | 26.8 | 25.1 | 26.7 | 25.4 |
|
| 104 |
+
| Ziya-LLaMA-13B-Pretrain-v1 | 30.2 | 27.8 | 34.3 | 32.0 | 29.0 |
|
| 105 |
+
| **XVERSE-13B** | **54.7** | **45.6** | **66.2** | **58.3** | **56.9** |
|
| 106 |
+
|
| 107 |
+
### Loading with Transformers
|
| 108 |
+
|
| 109 |
+
可通过以下代码加载 XVERSE-13B 模型进行推理：
|
| 110 |
+
|
| 111 |
+
The XVERSE-13B model can be loaded for inference using the following code:
|
| 112 |
+
|
| 113 |
+
```python
|
| 114 |
+
>>> from transformers import AutoTokenizer, AutoModelForCausalLM
|
| 115 |
+
>>> tokenizer = AutoTokenizer.from_pretrained("xverse/XVERSE-13B")
|
| 116 |
+
>>> model = AutoModelForCausalLM.from_pretrained("xverse/XVERSE-13B", trust_remote_code=True).half().cuda()
|
| 117 |
+
>>> model = model.eval()
|
| 118 |
+
>>> inputs = tokenizer('北京的景点：故宫、天坛、万里长城等。\n深圳的景点：', return_tensors='pt').input_ids
|
| 119 |
+
>>> inputs = inputs.cuda()
|
| 120 |
+
>>> generated_ids = model.generate(inputs, max_new_tokens=64, eos_token_id=tokenizer.eos_token_id)
|
| 121 |
+
>>> print(tokenizer.batch_decode(generated_ids, skip_special_tokens=True))
|
| 122 |
+
```
|
| 123 |
+
|
| 124 |
+
更多有关相关细节，包括文本生成demo和环境依赖，请参考我们的[Github](https://github.com/xverse-ai/XVERSE-13B)。
|
| 125 |
+
|
| 126 |
+
For more details, including the demo of text generation and environmental dependencies, please refer to our [Github](https://github.com/xverse-ai/XVERSE-13B).
|
| 127 |
+
|
| 128 |
+
|
| 129 |
+
## 局限性与免责申明
|
| 130 |
+
|
| 131 |
+
XVERSE-13B 与其他所有 LLM 一样，在某些情况下可能会产生不准确、有偏见或其他令人反感的内容。因此，请谨慎使用模型生成的内容，请勿将生成的有害内容进行传播，在部署任何 XVERSE-13B 的应用之前，开发人员应根据其具体应用对模型进行安全测试和调优。
|
| 132 |
+
|
| 133 |
+
我们强烈警告不要将 XVERSE-13B 模型用于制造或传播有害信息，或进行任何可能损害公众、国家、社会安全或违反法规的活动。如果使用 XVERSE-13B 模型产生任何问题，无论是数据安全问题、公共舆论风险，还是模型被误解、滥用、传播或不合规使用所引发的任何风险和问题，我们将不承担任何责任。
|
| 134 |
+
|
| 135 |
+
## Limitations and Disclaimer
|
| 136 |
+
|
| 137 |
+
Like all other Large Language Models (LLMs), XVERSE-13B may produce inaccurate, biased, or otherwise offensive content under certain circumstances. Therefore, please use the content generated by the model with caution and refrain from disseminating harmful content. Before deploying any application of XVERSE-13B, developers should conduct safety tests and optimization of the model according to its specific application.
|
| 138 |
+
|
| 139 |
+
We strongly warn against the use of the XVERSE-13B model for producing or spreading harmful information, or conducting any activities that might harm the public, national, or social security, or violate regulations. We assume no responsibility for any problems arising from the use of the XVERSE-13B model, whether it be data security issues, public opinion risks, or any risks and issues caused by misunderstanding, misuse, dissemination, or non-compliance with the model.
|
| 140 |
+
|
| 141 |
+
## 模型开源协议
|
| 142 |
+
|
| 143 |
+
使用本仓库的源码需要遵循 [Apache-2.0](https://github.com/xverse-ai/XVERSE-13B/blob/main/LICENSE) 开源协议，使用 XVERSE-13B 的模型权重则需要遵循[模型许可协议](MODEL_LICENSE.pdf)。
|
| 144 |
+
|
| 145 |
+
XVERSE-13B 模型权重对学术研究**完全开放**，并且支持**免费商用**，商用需申请商业使用授权，可以发送邮件到 <opensource@xverse.cn> 进行申请。
|
| 146 |
+
|
| 147 |
+
## Open Source License
|
| 148 |
+
|
| 149 |
+
The use of the source code in this repository must follow the [Apache-2.0](https://github.com/xverse-ai/XVERSE-13B/blob/main/LICENSE) open-source license, while the use of the model weights of XVERSE-13B needs to adhere to the [Model License Agreement](MODEL_LICENSE.pdf).
|
| 150 |
+
|
| 151 |
+
The XVERSE-13B model weights are **fully open** to academic research and support **free commercial use**. Commercial use requires an application for a commercial use license by sending an email to <opensource@xverse.cn>.
|
| 152 |
+
|
config.json
ADDED
|
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"XverseForCausalLM"
|
| 4 |
+
],
|
| 5 |
+
"auto_map": {
|
| 6 |
+
"AutoConfig": "configuration_xverse.XverseConfig",
|
| 7 |
+
"AutoModelForCausalLM": "modeling_xverse.XverseForCausalLM"
|
| 8 |
+
},
|
| 9 |
+
"pad_token_id": 1,
|
| 10 |
+
"bos_token_id": 2,
|
| 11 |
+
"eos_token_id": 3,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 5120,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 13824,
|
| 16 |
+
"max_position_embeddings": 8192,
|
| 17 |
+
"model_type": "xverse",
|
| 18 |
+
"num_attention_heads": 40,
|
| 19 |
+
"num_hidden_layers": 40,
|
| 20 |
+
"rms_norm_eps": 1e-06,
|
| 21 |
+
"tie_word_embeddings": false,
|
| 22 |
+
"torch_dtype": "float16",
|
| 23 |
+
"transformers_version": "4.28.1",
|
| 24 |
+
"use_cache": true,
|
| 25 |
+
"vocab_size": 100278
|
| 26 |
+
}
|
| 27 |
+
|
configuration_xverse.py
ADDED
|
@@ -0,0 +1,121 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# coding=utf-8
|
| 2 |
+
# Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
|
| 3 |
+
#
|
| 4 |
+
# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
|
| 5 |
+
# and OPT implementations in this library. It has been modified from its
|
| 6 |
+
# original forms to accommodate minor architectural differences compared
|
| 7 |
+
# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
|
| 8 |
+
#
|
| 9 |
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
| 10 |
+
# you may not use this file except in compliance with the License.
|
| 11 |
+
# You may obtain a copy of the License at
|
| 12 |
+
#
|
| 13 |
+
# http://www.apache.org/licenses/LICENSE-2.0
|
| 14 |
+
#
|
| 15 |
+
# Unless required by applicable law or agreed to in writing, software
|
| 16 |
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
| 17 |
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
| 18 |
+
# See the License for the specific language governing permissions and
|
| 19 |
+
# limitations under the License.
|
| 20 |
+
""" XVERSE model configuration"""
|
| 21 |
+
|
| 22 |
+
from transformers.configuration_utils import PretrainedConfig
|
| 23 |
+
from transformers.utils import logging
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
logger = logging.get_logger(__name__)
|
| 27 |
+
|
| 28 |
+
XVERSE_PRETRAINED_CONFIG_ARCHIVE_MAP = {}
|
| 29 |
+
|
| 30 |
+
|
| 31 |
+
class XverseConfig(PretrainedConfig):
|
| 32 |
+
r"""
|
| 33 |
+
This is the configuration class to store the configuration of a [`XverseModel`]. It is used to instantiate an Xverse
|
| 34 |
+
model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
|
| 35 |
+
defaults will yield a similar configuration to that of the XVERSE-13B.
|
| 36 |
+
|
| 37 |
+
Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
|
| 38 |
+
documentation from [`PretrainedConfig`] for more information.
|
| 39 |
+
|
| 40 |
+
|
| 41 |
+
Args:
|
| 42 |
+
vocab_size (`int`, *optional*, defaults to 100278):
|
| 43 |
+
Vocabulary size of the XVERSE model. Defines the number of different tokens that can be represented by the
|
| 44 |
+
`inputs_ids` passed when calling [`XverseModel`]
|
| 45 |
+
hidden_size (`int`, *optional*, defaults to 5120):
|
| 46 |
+
Dimension of the hidden representations.
|
| 47 |
+
intermediate_size (`int`, *optional*, defaults to 13824):
|
| 48 |
+
Dimension of the MLP representations.
|
| 49 |
+
num_hidden_layers (`int`, *optional*, defaults to 40):
|
| 50 |
+
Number of hidden layers in the Transformer encoder.
|
| 51 |
+
num_attention_heads (`int`, *optional*, defaults to 40):
|
| 52 |
+
Number of attention heads for each attention layer in the Transformer encoder.
|
| 53 |
+
hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
|
| 54 |
+
The non-linear activation function (function or string) in the decoder.
|
| 55 |
+
max_position_embeddings (`int`, *optional*, defaults to 8192):
|
| 56 |
+
The maximum sequence length that this model might ever be used with. Typically set this to something large
|
| 57 |
+
just in case (e.g., 512 or 1024 or 2048).
|
| 58 |
+
initializer_range (`float`, *optional*, defaults to 0.02):
|
| 59 |
+
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
|
| 60 |
+
rms_norm_eps (`float`, *optional*, defaults to 1e-6):
|
| 61 |
+
The epsilon used by the rms normalization layers.
|
| 62 |
+
use_cache (`bool`, *optional*, defaults to `True`):
|
| 63 |
+
Whether or not the model should return the last key/values attentions (not used by all models). Only
|
| 64 |
+
relevant if `config.is_decoder=True`.
|
| 65 |
+
tie_word_embeddings(`bool`, *optional*, defaults to `False`):
|
| 66 |
+
Whether to tie weight embeddings
|
| 67 |
+
|
| 68 |
+
Example:
|
| 69 |
+
|
| 70 |
+
```python
|
| 71 |
+
>>> from transformers import XverseModel, XverseConfig
|
| 72 |
+
|
| 73 |
+
>>> # Initializing a Xverse XVERSE-13B style configuration
|
| 74 |
+
>>> configuration = XverseConfig()
|
| 75 |
+
|
| 76 |
+
>>> # Initializing a model from the XVERSE-13B style configuration
|
| 77 |
+
>>> model = XverseModel(configuration)
|
| 78 |
+
|
| 79 |
+
>>> # Accessing the model configuration
|
| 80 |
+
>>> configuration = model.config
|
| 81 |
+
```"""
|
| 82 |
+
model_type = "xverse"
|
| 83 |
+
keys_to_ignore_at_inference = ["past_key_values"]
|
| 84 |
+
|
| 85 |
+
def __init__(
|
| 86 |
+
self,
|
| 87 |
+
vocab_size=100278,
|
| 88 |
+
hidden_size=5120,
|
| 89 |
+
intermediate_size=13824,
|
| 90 |
+
num_hidden_layers=40,
|
| 91 |
+
num_attention_heads=40,
|
| 92 |
+
hidden_act="silu",
|
| 93 |
+
max_position_embeddings=8192,
|
| 94 |
+
initializer_range=0.02,
|
| 95 |
+
rms_norm_eps=1e-6,
|
| 96 |
+
use_cache=True,
|
| 97 |
+
pad_token_id=None,
|
| 98 |
+
bos_token_id=1,
|
| 99 |
+
eos_token_id=2,
|
| 100 |
+
tie_word_embeddings=False,
|
| 101 |
+
**kwargs,
|
| 102 |
+
):
|
| 103 |
+
self.vocab_size = vocab_size
|
| 104 |
+
self.max_position_embeddings = max_position_embeddings
|
| 105 |
+
self.hidden_size = hidden_size
|
| 106 |
+
self.intermediate_size = intermediate_size
|
| 107 |
+
self.num_hidden_layers = num_hidden_layers
|
| 108 |
+
self.num_attention_heads = num_attention_heads
|
| 109 |
+
|
| 110 |
+
self.hidden_act = hidden_act
|
| 111 |
+
self.initializer_range = initializer_range
|
| 112 |
+
self.rms_norm_eps = rms_norm_eps
|
| 113 |
+
self.use_cache = use_cache
|
| 114 |
+
|
| 115 |
+
super().__init__(
|
| 116 |
+
pad_token_id=pad_token_id,
|
| 117 |
+
bos_token_id=bos_token_id,
|
| 118 |
+
eos_token_id=eos_token_id,
|
| 119 |
+
tie_word_embeddings=tie_word_embeddings,
|
| 120 |
+
**kwargs,
|
| 121 |
+
)
|
modeling_xverse.py
ADDED
|
@@ -0,0 +1,767 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# coding=utf-8
|
| 2 |
+
# Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
|
| 3 |
+
#
|
| 4 |
+
# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
|
| 5 |
+
# and OPT implementations in this library. It has been modified from its
|
| 6 |
+
# original forms to accommodate minor architectural differences compared
|
| 7 |
+
# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
|
| 8 |
+
#
|
| 9 |
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
| 10 |
+
# you may not use this file except in compliance with the License.
|
| 11 |
+
# You may obtain a copy of the License at
|
| 12 |
+
#
|
| 13 |
+
# http://www.apache.org/licenses/LICENSE-2.0
|
| 14 |
+
#
|
| 15 |
+
# Unless required by applicable law or agreed to in writing, software
|
| 16 |
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
| 17 |
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
| 18 |
+
# See the License for the specific language governing permissions and
|
| 19 |
+
# limitations under the License.
|
| 20 |
+
""" PyTorch XVERSE model."""
|
| 21 |
+
import math
|
| 22 |
+
from typing import List, Optional, Tuple, Union
|
| 23 |
+
|
| 24 |
+
import torch
|
| 25 |
+
import torch.nn.functional as F
|
| 26 |
+
import torch.utils.checkpoint
|
| 27 |
+
from torch import nn
|
| 28 |
+
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
|
| 29 |
+
|
| 30 |
+
from transformers.activations import ACT2FN
|
| 31 |
+
from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
|
| 32 |
+
from transformers.modeling_utils import PreTrainedModel
|
| 33 |
+
from transformers.utils import add_start_docstrings, add_start_docstrings_to_model_forward, logging, replace_return_docstrings
|
| 34 |
+
from .configuration_xverse import XverseConfig
|
| 35 |
+
|
| 36 |
+
|
| 37 |
+
logger = logging.get_logger(__name__)
|
| 38 |
+
|
| 39 |
+
_CONFIG_FOR_DOC = "XverseConfig"
|
| 40 |
+
|
| 41 |
+
|
| 42 |
+
# Copied from transformers.models.bart.modeling_bart._make_causal_mask
|
| 43 |
+
def _make_causal_mask(
|
| 44 |
+
input_ids_shape: torch.Size, dtype: torch.dtype, device: torch.device, past_key_values_length: int = 0
|
| 45 |
+
):
|
| 46 |
+
"""
|
| 47 |
+
Make causal mask used for bi-directional self-attention.
|
| 48 |
+
"""
|
| 49 |
+
bsz, tgt_len = input_ids_shape
|
| 50 |
+
mask = torch.full((tgt_len, tgt_len), torch.tensor(torch.finfo(dtype).min, device=device), device=device)
|
| 51 |
+
mask_cond = torch.arange(mask.size(-1), device=device)
|
| 52 |
+
mask.masked_fill_(mask_cond < (mask_cond + 1).view(mask.size(-1), 1), 0)
|
| 53 |
+
mask = mask.to(dtype)
|
| 54 |
+
|
| 55 |
+
if past_key_values_length > 0:
|
| 56 |
+
mask = torch.cat([torch.zeros(tgt_len, past_key_values_length, dtype=dtype, device=device), mask], dim=-1)
|
| 57 |
+
return mask[None, None, :, :].expand(bsz, 1, tgt_len, tgt_len + past_key_values_length)
|
| 58 |
+
|
| 59 |
+
|
| 60 |
+
# Copied from transformers.models.bart.modeling_bart._expand_mask
|
| 61 |
+
def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
|
| 62 |
+
"""
|
| 63 |
+
Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
|
| 64 |
+
"""
|
| 65 |
+
bsz, src_len = mask.size()
|
| 66 |
+
tgt_len = tgt_len if tgt_len is not None else src_len
|
| 67 |
+
|
| 68 |
+
expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)
|
| 69 |
+
|
| 70 |
+
inverted_mask = 1.0 - expanded_mask
|
| 71 |
+
|
| 72 |
+
return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)
|
| 73 |
+
|
| 74 |
+
|
| 75 |
+
class XverseRMSNorm(nn.Module):
|
| 76 |
+
def __init__(self, hidden_size, eps=1e-6):
|
| 77 |
+
"""
|
| 78 |
+
XverseRMSNorm is equivalent to T5LayerNorm
|
| 79 |
+
"""
|
| 80 |
+
super().__init__()
|
| 81 |
+
self.weight = nn.Parameter(torch.ones(hidden_size))
|
| 82 |
+
self.variance_epsilon = eps
|
| 83 |
+
|
| 84 |
+
def forward(self, hidden_states):
|
| 85 |
+
input_dtype = hidden_states.dtype
|
| 86 |
+
variance = hidden_states.to(torch.float32).pow(2).mean(-1, keepdim=True)
|
| 87 |
+
hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
|
| 88 |
+
|
| 89 |
+
return (self.weight * hidden_states).to(input_dtype)
|
| 90 |
+
|
| 91 |
+
|
| 92 |
+
class XverseRotaryEmbedding(torch.nn.Module):
|
| 93 |
+
def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None):
|
| 94 |
+
super().__init__()
|
| 95 |
+
inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2).float().to(device) / dim))
|
| 96 |
+
self.register_buffer("inv_freq", inv_freq)
|
| 97 |
+
|
| 98 |
+
# Build here to make `torch.jit.trace` work.
|
| 99 |
+
self.max_seq_len_cached = max_position_embeddings
|
| 100 |
+
t = torch.arange(self.max_seq_len_cached, device=self.inv_freq.device, dtype=self.inv_freq.dtype)
|
| 101 |
+
freqs = torch.einsum("i,j->ij", t, self.inv_freq)
|
| 102 |
+
# Different from paper, but it uses a different permutation in order to obtain the same calculation
|
| 103 |
+
emb = torch.cat((freqs, freqs), dim=-1)
|
| 104 |
+
self.register_buffer("cos_cached", emb.cos()[None, None, :, :], persistent=False)
|
| 105 |
+
self.register_buffer("sin_cached", emb.sin()[None, None, :, :], persistent=False)
|
| 106 |
+
|
| 107 |
+
def forward(self, x, seq_len=None):
|
| 108 |
+
# x: [bs, num_attention_heads, seq_len, head_size]
|
| 109 |
+
# This `if` block is unlikely to be run after we build sin/cos in `__init__`. Keep the logic here just in case.
|
| 110 |
+
if seq_len > self.max_seq_len_cached:
|
| 111 |
+
self.max_seq_len_cached = seq_len
|
| 112 |
+
t = torch.arange(self.max_seq_len_cached, device=x.device, dtype=self.inv_freq.dtype)
|
| 113 |
+
freqs = torch.einsum("i,j->ij", t, self.inv_freq)
|
| 114 |
+
# Different from paper, but it uses a different permutation in order to obtain the same calculation
|
| 115 |
+
emb = torch.cat((freqs, freqs), dim=-1).to(x.device)
|
| 116 |
+
self.register_buffer("cos_cached", emb.cos()[None, None, :, :], persistent=False)
|
| 117 |
+
self.register_buffer("sin_cached", emb.sin()[None, None, :, :], persistent=False)
|
| 118 |
+
return (
|
| 119 |
+
self.cos_cached[:, :, :seq_len, ...].to(dtype=x.dtype),
|
| 120 |
+
self.sin_cached[:, :, :seq_len, ...].to(dtype=x.dtype),
|
| 121 |
+
)
|
| 122 |
+
|
| 123 |
+
|
| 124 |
+
def rotate_half(x):
|
| 125 |
+
"""Rotates half the hidden dims of the input."""
|
| 126 |
+
x1 = x[..., : x.shape[-1] // 2]
|
| 127 |
+
x2 = x[..., x.shape[-1] // 2 :]
|
| 128 |
+
return torch.cat((-x2, x1), dim=-1)
|
| 129 |
+
|
| 130 |
+
|
| 131 |
+
def apply_rotary_pos_emb(q, k, cos, sin, position_ids):
|
| 132 |
+
# The first two dimensions of cos and sin are always 1, so we can `squeeze` them.
|
| 133 |
+
cos = cos.squeeze(1).squeeze(0) # [seq_len, dim]
|
| 134 |
+
sin = sin.squeeze(1).squeeze(0) # [seq_len, dim]
|
| 135 |
+
cos = cos[position_ids].unsqueeze(1) # [bs, 1, seq_len, dim]
|
| 136 |
+
sin = sin[position_ids].unsqueeze(1) # [bs, 1, seq_len, dim]
|
| 137 |
+
q_embed = (q * cos) + (rotate_half(q) * sin)
|
| 138 |
+
k_embed = (k * cos) + (rotate_half(k) * sin)
|
| 139 |
+
return q_embed, k_embed
|
| 140 |
+
|
| 141 |
+
|
| 142 |
+
class XverseMLP(nn.Module):
|
| 143 |
+
def __init__(
|
| 144 |
+
self,
|
| 145 |
+
hidden_size: int,
|
| 146 |
+
intermediate_size: int,
|
| 147 |
+
hidden_act: str,
|
| 148 |
+
):
|
| 149 |
+
super().__init__()
|
| 150 |
+
self.gate_proj = nn.Linear(hidden_size, intermediate_size, bias=False)
|
| 151 |
+
self.down_proj = nn.Linear(intermediate_size, hidden_size, bias=False)
|
| 152 |
+
self.up_proj = nn.Linear(hidden_size, intermediate_size, bias=False)
|
| 153 |
+
self.act_fn = ACT2FN[hidden_act]
|
| 154 |
+
|
| 155 |
+
def forward(self, x):
|
| 156 |
+
return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
|
| 157 |
+
|
| 158 |
+
|
| 159 |
+
class XverseAttention(nn.Module):
|
| 160 |
+
"""Multi-headed attention from 'Attention Is All You Need' paper"""
|
| 161 |
+
|
| 162 |
+
def __init__(self, config: XverseConfig):
|
| 163 |
+
super().__init__()
|
| 164 |
+
self.config = config
|
| 165 |
+
self.hidden_size = config.hidden_size
|
| 166 |
+
self.num_heads = config.num_attention_heads
|
| 167 |
+
self.head_dim = self.hidden_size // self.num_heads
|
| 168 |
+
self.max_position_embeddings = config.max_position_embeddings
|
| 169 |
+
|
| 170 |
+
if (self.head_dim * self.num_heads) != self.hidden_size:
|
| 171 |
+
raise ValueError(
|
| 172 |
+
f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
|
| 173 |
+
f" and `num_heads`: {self.num_heads})."
|
| 174 |
+
)
|
| 175 |
+
self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=False)
|
| 176 |
+
self.k_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=False)
|
| 177 |
+
self.v_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=False)
|
| 178 |
+
self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=False)
|
| 179 |
+
self.rotary_emb = XverseRotaryEmbedding(self.head_dim, max_position_embeddings=self.max_position_embeddings)
|
| 180 |
+
|
| 181 |
+
def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
|
| 182 |
+
return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
|
| 183 |
+
|
| 184 |
+
def forward(
|
| 185 |
+
self,
|
| 186 |
+
hidden_states: torch.Tensor,
|
| 187 |
+
attention_mask: Optional[torch.Tensor] = None,
|
| 188 |
+
position_ids: Optional[torch.LongTensor] = None,
|
| 189 |
+
past_key_value: Optional[Tuple[torch.Tensor]] = None,
|
| 190 |
+
output_attentions: bool = False,
|
| 191 |
+
use_cache: bool = False,
|
| 192 |
+
) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
|
| 193 |
+
bsz, q_len, _ = hidden_states.size()
|
| 194 |
+
|
| 195 |
+
query_states = self.q_proj(hidden_states).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
|
| 196 |
+
key_states = self.k_proj(hidden_states).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
|
| 197 |
+
value_states = self.v_proj(hidden_states).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
|
| 198 |
+
|
| 199 |
+
kv_seq_len = key_states.shape[-2]
|
| 200 |
+
if past_key_value is not None:
|
| 201 |
+
kv_seq_len += past_key_value[0].shape[-2]
|
| 202 |
+
cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
|
| 203 |
+
query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
|
| 204 |
+
# [bsz, nh, t, hd]
|
| 205 |
+
|
| 206 |
+
if past_key_value is not None:
|
| 207 |
+
# reuse k, v, self_attention
|
| 208 |
+
key_states = torch.cat([past_key_value[0], key_states], dim=2)
|
| 209 |
+
value_states = torch.cat([past_key_value[1], value_states], dim=2)
|
| 210 |
+
|
| 211 |
+
past_key_value = (key_states, value_states) if use_cache else None
|
| 212 |
+
|
| 213 |
+
attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim)
|
| 214 |
+
|
| 215 |
+
if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
|
| 216 |
+
raise ValueError(
|
| 217 |
+
f"Attention weights should be of size {(bsz, self.num_heads, q_len, kv_seq_len)}, but is"
|
| 218 |
+
f" {attn_weights.size()}"
|
| 219 |
+
)
|
| 220 |
+
|
| 221 |
+
if attention_mask is not None:
|
| 222 |
+
if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
|
| 223 |
+
raise ValueError(
|
| 224 |
+
f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
|
| 225 |
+
)
|
| 226 |
+
attn_weights = attn_weights + attention_mask
|
| 227 |
+
attn_weights = torch.max(
|
| 228 |
+
attn_weights, torch.tensor(torch.finfo(attn_weights.dtype).min, device=attn_weights.device)
|
| 229 |
+
)
|
| 230 |
+
|
| 231 |
+
# upcast attention to fp32
|
| 232 |
+
attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
|
| 233 |
+
attn_output = torch.matmul(attn_weights, value_states)
|
| 234 |
+
|
| 235 |
+
if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
|
| 236 |
+
raise ValueError(
|
| 237 |
+
f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
|
| 238 |
+
f" {attn_output.size()}"
|
| 239 |
+
)
|
| 240 |
+
|
| 241 |
+
attn_output = attn_output.transpose(1, 2)
|
| 242 |
+
attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
|
| 243 |
+
|
| 244 |
+
attn_output = self.o_proj(attn_output)
|
| 245 |
+
|
| 246 |
+
if not output_attentions:
|
| 247 |
+
attn_weights = None
|
| 248 |
+
|
| 249 |
+
return attn_output, attn_weights, past_key_value
|
| 250 |
+
|
| 251 |
+
|
| 252 |
+
class XverseDecoderLayer(nn.Module):
|
| 253 |
+
def __init__(self, config: XverseConfig):
|
| 254 |
+
super().__init__()
|
| 255 |
+
self.hidden_size = config.hidden_size
|
| 256 |
+
self.self_attn = XverseAttention(config=config)
|
| 257 |
+
self.mlp = XverseMLP(
|
| 258 |
+
hidden_size=self.hidden_size,
|
| 259 |
+
intermediate_size=config.intermediate_size,
|
| 260 |
+
hidden_act=config.hidden_act,
|
| 261 |
+
)
|
| 262 |
+
self.input_layernorm = XverseRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
|
| 263 |
+
self.post_attention_layernorm = XverseRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
|
| 264 |
+
|
| 265 |
+
def forward(
|
| 266 |
+
self,
|
| 267 |
+
hidden_states: torch.Tensor,
|
| 268 |
+
attention_mask: Optional[torch.Tensor] = None,
|
| 269 |
+
position_ids: Optional[torch.LongTensor] = None,
|
| 270 |
+
past_key_value: Optional[Tuple[torch.Tensor]] = None,
|
| 271 |
+
output_attentions: Optional[bool] = False,
|
| 272 |
+
use_cache: Optional[bool] = False,
|
| 273 |
+
) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
|
| 274 |
+
"""
|
| 275 |
+
Args:
|
| 276 |
+
hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
|
| 277 |
+
attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
|
| 278 |
+
`(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
|
| 279 |
+
output_attentions (`bool`, *optional*):
|
| 280 |
+
Whether or not to return the attentions tensors of all attention layers. See `attentions` under
|
| 281 |
+
returned tensors for more detail.
|
| 282 |
+
use_cache (`bool`, *optional*):
|
| 283 |
+
If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
|
| 284 |
+
(see `past_key_values`).
|
| 285 |
+
past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
|
| 286 |
+
"""
|
| 287 |
+
|
| 288 |
+
residual = hidden_states
|
| 289 |
+
|
| 290 |
+
hidden_states = self.input_layernorm(hidden_states)
|
| 291 |
+
|
| 292 |
+
# Self Attention
|
| 293 |
+
hidden_states, self_attn_weights, present_key_value = self.self_attn(
|
| 294 |
+
hidden_states=hidden_states,
|
| 295 |
+
attention_mask=attention_mask,
|
| 296 |
+
position_ids=position_ids,
|
| 297 |
+
past_key_value=past_key_value,
|
| 298 |
+
output_attentions=output_attentions,
|
| 299 |
+
use_cache=use_cache,
|
| 300 |
+
)
|
| 301 |
+
hidden_states = residual + hidden_states
|
| 302 |
+
|
| 303 |
+
# Fully Connected
|
| 304 |
+
residual = hidden_states
|
| 305 |
+
hidden_states = self.post_attention_layernorm(hidden_states)
|
| 306 |
+
hidden_states = self.mlp(hidden_states)
|
| 307 |
+
hidden_states = residual + hidden_states
|
| 308 |
+
|
| 309 |
+
outputs = (hidden_states,)
|
| 310 |
+
|
| 311 |
+
if output_attentions:
|
| 312 |
+
outputs += (self_attn_weights,)
|
| 313 |
+
|
| 314 |
+
if use_cache:
|
| 315 |
+
outputs += (present_key_value,)
|
| 316 |
+
|
| 317 |
+
return outputs
|
| 318 |
+
|
| 319 |
+
|
| 320 |
+
XVERSE_START_DOCSTRING = r"""
|
| 321 |
+
This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
|
| 322 |
+
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
|
| 323 |
+
etc.)
|
| 324 |
+
|
| 325 |
+
This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
|
| 326 |
+
Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
|
| 327 |
+
and behavior.
|
| 328 |
+
|
| 329 |
+
Parameters:
|
| 330 |
+
config ([`XverseConfig`]):
|
| 331 |
+
Model configuration class with all the parameters of the model. Initializing with a config file does not
|
| 332 |
+
load the weights associated with the model, only the configuration. Check out the
|
| 333 |
+
[`~PreTrainedModel.from_pretrained`] method to load the model weights.
|
| 334 |
+
"""
|
| 335 |
+
|
| 336 |
+
|
| 337 |
+
@add_start_docstrings(
|
| 338 |
+
"The bare Xverse Model outputting raw hidden-states without any specific head on top.",
|
| 339 |
+
XVERSE_START_DOCSTRING,
|
| 340 |
+
)
|
| 341 |
+
class XversePreTrainedModel(PreTrainedModel):
|
| 342 |
+
config_class = XverseConfig
|
| 343 |
+
base_model_prefix = "model"
|
| 344 |
+
supports_gradient_checkpointing = True
|
| 345 |
+
_no_split_modules = ["XverseDecoderLayer"]
|
| 346 |
+
_skip_keys_device_placement = "past_key_values"
|
| 347 |
+
_keys_to_ignore_on_load_unexpected = [r"decoder\.version"]
|
| 348 |
+
|
| 349 |
+
def _init_weights(self, module):
|
| 350 |
+
std = self.config.initializer_range
|
| 351 |
+
if isinstance(module, nn.Linear):
|
| 352 |
+
module.weight.data.normal_(mean=0.0, std=std)
|
| 353 |
+
if module.bias is not None:
|
| 354 |
+
module.bias.data.zero_()
|
| 355 |
+
elif isinstance(module, nn.Embedding):
|
| 356 |
+
module.weight.data.normal_(mean=0.0, std=std)
|
| 357 |
+
if module.padding_idx is not None:
|
| 358 |
+
module.weight.data[module.padding_idx].zero_()
|
| 359 |
+
|
| 360 |
+
def _set_gradient_checkpointing(self, module, value=False):
|
| 361 |
+
if isinstance(module, XverseModel):
|
| 362 |
+
module.gradient_checkpointing = value
|
| 363 |
+
|
| 364 |
+
|
| 365 |
+
XVERSE_INPUTS_DOCSTRING = r"""
|
| 366 |
+
Args:
|
| 367 |
+
input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
|
| 368 |
+
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
|
| 369 |
+
it.
|
| 370 |
+
|
| 371 |
+
Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
|
| 372 |
+
[`PreTrainedTokenizer.__call__`] for details.
|
| 373 |
+
|
| 374 |
+
[What are input IDs?](../glossary#input-ids)
|
| 375 |
+
attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
|
| 376 |
+
Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
|
| 377 |
+
|
| 378 |
+
- 1 for tokens that are **not masked**,
|
| 379 |
+
- 0 for tokens that are **masked**.
|
| 380 |
+
|
| 381 |
+
[What are attention masks?](../glossary#attention-mask)
|
| 382 |
+
|
| 383 |
+
Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
|
| 384 |
+
[`PreTrainedTokenizer.__call__`] for details.
|
| 385 |
+
|
| 386 |
+
If `past_key_values` is used, optionally only the last `decoder_input_ids` have to be input (see
|
| 387 |
+
`past_key_values`).
|
| 388 |
+
|
| 389 |
+
If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`]
|
| 390 |
+
and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more
|
| 391 |
+
information on the default strategy.
|
| 392 |
+
|
| 393 |
+
- 1 indicates the head is **not masked**,
|
| 394 |
+
- 0 indicates the head is **masked**.
|
| 395 |
+
position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
|
| 396 |
+
Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
|
| 397 |
+
config.n_positions - 1]`.
|
| 398 |
+
|
| 399 |
+
[What are position IDs?](../glossary#position-ids)
|
| 400 |
+
past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
|
| 401 |
+
Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
|
| 402 |
+
`(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of shape
|
| 403 |
+
`(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)`.
|
| 404 |
+
|
| 405 |
+
Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
|
| 406 |
+
blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
|
| 407 |
+
|
| 408 |
+
If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those that
|
| 409 |
+
don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all
|
| 410 |
+
`decoder_input_ids` of shape `(batch_size, sequence_length)`.
|
| 411 |
+
inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
|
| 412 |
+
Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
|
| 413 |
+
is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
|
| 414 |
+
model's internal embedding lookup matrix.
|
| 415 |
+
use_cache (`bool`, *optional*):
|
| 416 |
+
If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
|
| 417 |
+
`past_key_values`).
|
| 418 |
+
output_attentions (`bool`, *optional*):
|
| 419 |
+
Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
|
| 420 |
+
tensors for more detail.
|
| 421 |
+
output_hidden_states (`bool`, *optional*):
|
| 422 |
+
Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
|
| 423 |
+
more detail.
|
| 424 |
+
return_dict (`bool`, *optional*):
|
| 425 |
+
Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
|
| 426 |
+
"""
|
| 427 |
+
|
| 428 |
+
@add_start_docstrings(
|
| 429 |
+
"The bare Xverse Model outputting raw hidden-states without any specific head on top.",
|
| 430 |
+
XVERSE_START_DOCSTRING,
|
| 431 |
+
)
|
| 432 |
+
class XverseModel(XversePreTrainedModel):
|
| 433 |
+
"""
|
| 434 |
+
Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`XverseDecoderLayer`]
|
| 435 |
+
|
| 436 |
+
Args:
|
| 437 |
+
config: XverseConfig
|
| 438 |
+
"""
|
| 439 |
+
|
| 440 |
+
def __init__(self, config: XverseConfig):
|
| 441 |
+
super().__init__(config)
|
| 442 |
+
self.padding_idx = config.pad_token_id
|
| 443 |
+
self.vocab_size = config.vocab_size
|
| 444 |
+
|
| 445 |
+
self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
|
| 446 |
+
self.layers = nn.ModuleList([XverseDecoderLayer(config) for _ in range(config.num_hidden_layers)])
|
| 447 |
+
self.norm = XverseRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
|
| 448 |
+
|
| 449 |
+
self.gradient_checkpointing = False
|
| 450 |
+
# Initialize weights and apply final processing
|
| 451 |
+
self.post_init()
|
| 452 |
+
|
| 453 |
+
def get_input_embeddings(self):
|
| 454 |
+
return self.embed_tokens
|
| 455 |
+
|
| 456 |
+
def set_input_embeddings(self, value):
|
| 457 |
+
self.embed_tokens = value
|
| 458 |
+
|
| 459 |
+
# Copied from transformers.models.bart.modeling_bart.BartDecoder._prepare_decoder_attention_mask
|
| 460 |
+
def _prepare_decoder_attention_mask(self, attention_mask, input_shape, inputs_embeds, past_key_values_length):
|
| 461 |
+
# create causal mask
|
| 462 |
+
# [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
|
| 463 |
+
combined_attention_mask = None
|
| 464 |
+
if input_shape[-1] > 1:
|
| 465 |
+
combined_attention_mask = _make_causal_mask(
|
| 466 |
+
input_shape,
|
| 467 |
+
inputs_embeds.dtype,
|
| 468 |
+
device=inputs_embeds.device,
|
| 469 |
+
past_key_values_length=past_key_values_length,
|
| 470 |
+
)
|
| 471 |
+
|
| 472 |
+
if attention_mask is not None:
|
| 473 |
+
# [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
|
| 474 |
+
expanded_attn_mask = _expand_mask(attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]).to(
|
| 475 |
+
inputs_embeds.device
|
| 476 |
+
)
|
| 477 |
+
combined_attention_mask = (
|
| 478 |
+
expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask + combined_attention_mask
|
| 479 |
+
)
|
| 480 |
+
|
| 481 |
+
return combined_attention_mask
|
| 482 |
+
|
| 483 |
+
@add_start_docstrings_to_model_forward(XVERSE_INPUTS_DOCSTRING)
|
| 484 |
+
def forward(
|
| 485 |
+
self,
|
| 486 |
+
input_ids: torch.LongTensor = None,
|
| 487 |
+
attention_mask: Optional[torch.Tensor] = None,
|
| 488 |
+
position_ids: Optional[torch.LongTensor] = None,
|
| 489 |
+
past_key_values: Optional[List[torch.FloatTensor]] = None,
|
| 490 |
+
inputs_embeds: Optional[torch.FloatTensor] = None,
|
| 491 |
+
use_cache: Optional[bool] = None,
|
| 492 |
+
output_attentions: Optional[bool] = None,
|
| 493 |
+
output_hidden_states: Optional[bool] = None,
|
| 494 |
+
return_dict: Optional[bool] = None,
|
| 495 |
+
) -> Union[Tuple, BaseModelOutputWithPast]:
|
| 496 |
+
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
|
| 497 |
+
output_hidden_states = (
|
| 498 |
+
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
|
| 499 |
+
)
|
| 500 |
+
use_cache = use_cache if use_cache is not None else self.config.use_cache
|
| 501 |
+
|
| 502 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
| 503 |
+
|
| 504 |
+
# retrieve input_ids and inputs_embeds
|
| 505 |
+
if input_ids is not None and inputs_embeds is not None:
|
| 506 |
+
raise ValueError("You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time")
|
| 507 |
+
elif input_ids is not None:
|
| 508 |
+
batch_size, seq_length = input_ids.shape
|
| 509 |
+
elif inputs_embeds is not None:
|
| 510 |
+
batch_size, seq_length, _ = inputs_embeds.shape
|
| 511 |
+
else:
|
| 512 |
+
raise ValueError("You have to specify either decoder_input_ids or decoder_inputs_embeds")
|
| 513 |
+
|
| 514 |
+
seq_length_with_past = seq_length
|
| 515 |
+
past_key_values_length = 0
|
| 516 |
+
|
| 517 |
+
if past_key_values is not None:
|
| 518 |
+
past_key_values_length = past_key_values[0][0].shape[2]
|
| 519 |
+
seq_length_with_past = seq_length_with_past + past_key_values_length
|
| 520 |
+
|
| 521 |
+
if position_ids is None:
|
| 522 |
+
device = input_ids.device if input_ids is not None else inputs_embeds.device
|
| 523 |
+
position_ids = torch.arange(
|
| 524 |
+
past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
|
| 525 |
+
)
|
| 526 |
+
position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
|
| 527 |
+
else:
|
| 528 |
+
position_ids = position_ids.view(-1, seq_length).long()
|
| 529 |
+
|
| 530 |
+
if inputs_embeds is None:
|
| 531 |
+
inputs_embeds = self.embed_tokens(input_ids)
|
| 532 |
+
# embed positions
|
| 533 |
+
if attention_mask is None:
|
| 534 |
+
attention_mask = torch.ones(
|
| 535 |
+
(batch_size, seq_length_with_past), dtype=torch.bool, device=inputs_embeds.device
|
| 536 |
+
)
|
| 537 |
+
attention_mask = self._prepare_decoder_attention_mask(
|
| 538 |
+
attention_mask, (batch_size, seq_length), inputs_embeds, past_key_values_length
|
| 539 |
+
)
|
| 540 |
+
|
| 541 |
+
hidden_states = inputs_embeds
|
| 542 |
+
|
| 543 |
+
if self.gradient_checkpointing and self.training:
|
| 544 |
+
if use_cache:
|
| 545 |
+
logger.warning_once(
|
| 546 |
+
"`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
|
| 547 |
+
)
|
| 548 |
+
use_cache = False
|
| 549 |
+
|
| 550 |
+
# decoder layers
|
| 551 |
+
all_hidden_states = () if output_hidden_states else None
|
| 552 |
+
all_self_attns = () if output_attentions else None
|
| 553 |
+
next_decoder_cache = () if use_cache else None
|
| 554 |
+
|
| 555 |
+
for idx, decoder_layer in enumerate(self.layers):
|
| 556 |
+
if output_hidden_states:
|
| 557 |
+
all_hidden_states += (hidden_states,)
|
| 558 |
+
|
| 559 |
+
past_key_value = past_key_values[idx] if past_key_values is not None else None
|
| 560 |
+
|
| 561 |
+
if self.gradient_checkpointing and self.training:
|
| 562 |
+
|
| 563 |
+
def create_custom_forward(module):
|
| 564 |
+
def custom_forward(*inputs):
|
| 565 |
+
# None for past_key_value
|
| 566 |
+
return module(*inputs, output_attentions, None)
|
| 567 |
+
|
| 568 |
+
return custom_forward
|
| 569 |
+
|
| 570 |
+
layer_outputs = torch.utils.checkpoint.checkpoint(
|
| 571 |
+
create_custom_forward(decoder_layer),
|
| 572 |
+
hidden_states,
|
| 573 |
+
attention_mask,
|
| 574 |
+
position_ids,
|
| 575 |
+
None,
|
| 576 |
+
)
|
| 577 |
+
else:
|
| 578 |
+
layer_outputs = decoder_layer(
|
| 579 |
+
hidden_states,
|
| 580 |
+
attention_mask=attention_mask,
|
| 581 |
+
position_ids=position_ids,
|
| 582 |
+
past_key_value=past_key_value,
|
| 583 |
+
output_attentions=output_attentions,
|
| 584 |
+
use_cache=use_cache,
|
| 585 |
+
)
|
| 586 |
+
|
| 587 |
+
hidden_states = layer_outputs[0]
|
| 588 |
+
|
| 589 |
+
if use_cache:
|
| 590 |
+
next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
|
| 591 |
+
|
| 592 |
+
if output_attentions:
|
| 593 |
+
all_self_attns += (layer_outputs[1],)
|
| 594 |
+
|
| 595 |
+
hidden_states = self.norm(hidden_states)
|
| 596 |
+
|
| 597 |
+
# add hidden states from the last decoder layer
|
| 598 |
+
if output_hidden_states:
|
| 599 |
+
all_hidden_states += (hidden_states,)
|
| 600 |
+
|
| 601 |
+
next_cache = next_decoder_cache if use_cache else None
|
| 602 |
+
if not return_dict:
|
| 603 |
+
return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None)
|
| 604 |
+
return BaseModelOutputWithPast(
|
| 605 |
+
last_hidden_state=hidden_states,
|
| 606 |
+
past_key_values=next_cache,
|
| 607 |
+
hidden_states=all_hidden_states,
|
| 608 |
+
attentions=all_self_attns,
|
| 609 |
+
)
|
| 610 |
+
|
| 611 |
+
|
| 612 |
+
class XverseForCausalLM(XversePreTrainedModel):
|
| 613 |
+
_tied_weights_keys = ["lm_head.weight"]
|
| 614 |
+
|
| 615 |
+
def __init__(self, config):
|
| 616 |
+
super().__init__(config)
|
| 617 |
+
self.model = XverseModel(config)
|
| 618 |
+
|
| 619 |
+
self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
|
| 620 |
+
|
| 621 |
+
# Initialize weights and apply final processing
|
| 622 |
+
self.post_init()
|
| 623 |
+
|
| 624 |
+
def get_input_embeddings(self):
|
| 625 |
+
return self.model.embed_tokens
|
| 626 |
+
|
| 627 |
+
def set_input_embeddings(self, value):
|
| 628 |
+
self.model.embed_tokens = value
|
| 629 |
+
|
| 630 |
+
def get_output_embeddings(self):
|
| 631 |
+
return self.lm_head
|
| 632 |
+
|
| 633 |
+
def set_output_embeddings(self, new_embeddings):
|
| 634 |
+
self.lm_head = new_embeddings
|
| 635 |
+
|
| 636 |
+
def set_decoder(self, decoder):
|
| 637 |
+
self.model = decoder
|
| 638 |
+
|
| 639 |
+
def get_decoder(self):
|
| 640 |
+
return self.model
|
| 641 |
+
|
| 642 |
+
@add_start_docstrings_to_model_forward(XVERSE_INPUTS_DOCSTRING)
|
| 643 |
+
@replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
|
| 644 |
+
def forward(
|
| 645 |
+
self,
|
| 646 |
+
input_ids: torch.LongTensor = None,
|
| 647 |
+
attention_mask: Optional[torch.Tensor] = None,
|
| 648 |
+
position_ids: Optional[torch.LongTensor] = None,
|
| 649 |
+
past_key_values: Optional[List[torch.FloatTensor]] = None,
|
| 650 |
+
inputs_embeds: Optional[torch.FloatTensor] = None,
|
| 651 |
+
labels: Optional[torch.LongTensor] = None,
|
| 652 |
+
use_cache: Optional[bool] = None,
|
| 653 |
+
output_attentions: Optional[bool] = None,
|
| 654 |
+
output_hidden_states: Optional[bool] = None,
|
| 655 |
+
return_dict: Optional[bool] = None,
|
| 656 |
+
) -> Union[Tuple, CausalLMOutputWithPast]:
|
| 657 |
+
r"""
|
| 658 |
+
Args:
|
| 659 |
+
labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
|
| 660 |
+
Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
|
| 661 |
+
config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
|
| 662 |
+
(masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
|
| 663 |
+
|
| 664 |
+
Returns:
|
| 665 |
+
|
| 666 |
+
Example:
|
| 667 |
+
|
| 668 |
+
```python
|
| 669 |
+
>>> from transformers import AutoTokenizer, AutoModelForCausalLM
|
| 670 |
+
|
| 671 |
+
>>> model = AutoModelForCausalLM.from_pretrained(PATH_TO_CONVERTED_WEIGHTS, trust_remote_code=True)
|
| 672 |
+
>>> tokenizer = AutoTokenizer.from_pretrained(PATH_TO_CONVERTED_TOKENIZER)
|
| 673 |
+
|
| 674 |
+
>>> prompt = "Hey, are you conscious? Can you talk to me?"
|
| 675 |
+
>>> inputs = tokenizer(prompt, return_tensors="pt")
|
| 676 |
+
|
| 677 |
+
>>> # Generate
|
| 678 |
+
>>> generate_ids = model.generate(inputs.input_ids, max_length=30)
|
| 679 |
+
>>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
|
| 680 |
+
"Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
|
| 681 |
+
```"""
|
| 682 |
+
|
| 683 |
+
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
|
| 684 |
+
output_hidden_states = (
|
| 685 |
+
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
|
| 686 |
+
)
|
| 687 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
| 688 |
+
|
| 689 |
+
# decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
|
| 690 |
+
outputs = self.model(
|
| 691 |
+
input_ids=input_ids,
|
| 692 |
+
attention_mask=attention_mask,
|
| 693 |
+
position_ids=position_ids,
|
| 694 |
+
past_key_values=past_key_values,
|
| 695 |
+
inputs_embeds=inputs_embeds,
|
| 696 |
+
use_cache=use_cache,
|
| 697 |
+
output_attentions=output_attentions,
|
| 698 |
+
output_hidden_states=output_hidden_states,
|
| 699 |
+
return_dict=return_dict,
|
| 700 |
+
)
|
| 701 |
+
|
| 702 |
+
hidden_states = outputs[0]
|
| 703 |
+
logits = self.lm_head(hidden_states)
|
| 704 |
+
|
| 705 |
+
loss = None
|
| 706 |
+
if labels is not None:
|
| 707 |
+
# Shift so that tokens < n predict n
|
| 708 |
+
shift_logits = logits[..., :-1, :].contiguous()
|
| 709 |
+
shift_labels = labels[..., 1:].contiguous()
|
| 710 |
+
# Flatten the tokens
|
| 711 |
+
loss_fct = CrossEntropyLoss()
|
| 712 |
+
shift_logits = shift_logits.view(-1, self.config.vocab_size)
|
| 713 |
+
shift_labels = shift_labels.view(-1)
|
| 714 |
+
# Enable model parallelism
|
| 715 |
+
shift_labels = shift_labels.to(shift_logits.device)
|
| 716 |
+
loss = loss_fct(shift_logits, shift_labels)
|
| 717 |
+
|
| 718 |
+
if not return_dict:
|
| 719 |
+
output = (logits,) + outputs[1:]
|
| 720 |
+
return (loss,) + output if loss is not None else output
|
| 721 |
+
|
| 722 |
+
return CausalLMOutputWithPast(
|
| 723 |
+
loss=loss,
|
| 724 |
+
logits=logits,
|
| 725 |
+
past_key_values=outputs.past_key_values,
|
| 726 |
+
hidden_states=outputs.hidden_states,
|
| 727 |
+
attentions=outputs.attentions,
|
| 728 |
+
)
|
| 729 |
+
|
| 730 |
+
def prepare_inputs_for_generation(
|
| 731 |
+
self, input_ids, past_key_values=None, attention_mask=None, inputs_embeds=None, **kwargs
|
| 732 |
+
):
|
| 733 |
+
if past_key_values:
|
| 734 |
+
input_ids = input_ids[:, -1:]
|
| 735 |
+
|
| 736 |
+
position_ids = kwargs.get("position_ids", None)
|
| 737 |
+
if attention_mask is not None and position_ids is None:
|
| 738 |
+
# create position_ids on the fly for batch generation
|
| 739 |
+
position_ids = attention_mask.long().cumsum(-1) - 1
|
| 740 |
+
position_ids.masked_fill_(attention_mask == 0, 1)
|
| 741 |
+
if past_key_values:
|
| 742 |
+
position_ids = position_ids[:, -1].unsqueeze(-1)
|
| 743 |
+
|
| 744 |
+
# if `inputs_embeds` are passed, we only want to use them in the 1st generation step
|
| 745 |
+
if inputs_embeds is not None and past_key_values is None:
|
| 746 |
+
model_inputs = {"inputs_embeds": inputs_embeds}
|
| 747 |
+
else:
|
| 748 |
+
model_inputs = {"input_ids": input_ids}
|
| 749 |
+
|
| 750 |
+
model_inputs.update(
|
| 751 |
+
{
|
| 752 |
+
"position_ids": position_ids,
|
| 753 |
+
"past_key_values": past_key_values,
|
| 754 |
+
"use_cache": kwargs.get("use_cache"),
|
| 755 |
+
"attention_mask": attention_mask,
|
| 756 |
+
}
|
| 757 |
+
)
|
| 758 |
+
return model_inputs
|
| 759 |
+
|
| 760 |
+
@staticmethod
|
| 761 |
+
def _reorder_cache(past_key_values, beam_idx):
|
| 762 |
+
reordered_past = ()
|
| 763 |
+
for layer_past in past_key_values:
|
| 764 |
+
reordered_past += (
|
| 765 |
+
tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past),
|
| 766 |
+
)
|
| 767 |
+
return reordered_past
|
pytorch_model-00001-of-00003.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:18696855f373f92e774ac4ae12d80fe66b1c1e9acc7b98cd1029fef9e9d228d3
|
| 3 |
+
size 10065941445
|
pytorch_model-00002-of-00003.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:10bb394d79391a86c887edce8e7d2f1b921c60fcd9aefe89ee5fe6de9b370a0f
|
| 3 |
+
size 10009010414
|
pytorch_model-00003-of-00003.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:defb5f082a0b8924456672401a1d691cae47ce0e01d20e07255c8718d0d7b955
|
| 3 |
+
size 7512554533
|
pytorch_model.bin.index.json
ADDED
|
@@ -0,0 +1,410 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"metadata": {
|
| 3 |
+
"total_size": 27430067200
|
| 4 |
+
},
|
| 5 |
+
"weight_map": {
|
| 6 |
+
"lm_head.weight": "pytorch_model-00003-of-00003.bin",
|
| 7 |
+
"model.embed_tokens.weight": "pytorch_model-00001-of-00003.bin",
|
| 8 |
+
"model.layers.0.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 9 |
+
"model.layers.0.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 10 |
+
"model.layers.0.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 11 |
+
"model.layers.0.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 12 |
+
"model.layers.0.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 13 |
+
"model.layers.0.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 14 |
+
"model.layers.0.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 15 |
+
"model.layers.0.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 16 |
+
"model.layers.0.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
|
| 17 |
+
"model.layers.0.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 18 |
+
"model.layers.1.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 19 |
+
"model.layers.1.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 20 |
+
"model.layers.1.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 21 |
+
"model.layers.1.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 22 |
+
"model.layers.1.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 23 |
+
"model.layers.1.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 24 |
+
"model.layers.1.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 25 |
+
"model.layers.1.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 26 |
+
"model.layers.1.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
|
| 27 |
+
"model.layers.1.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 28 |
+
"model.layers.10.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 29 |
+
"model.layers.10.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 30 |
+
"model.layers.10.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 31 |
+
"model.layers.10.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 32 |
+
"model.layers.10.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 33 |
+
"model.layers.10.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 34 |
+
"model.layers.10.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 35 |
+
"model.layers.10.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 36 |
+
"model.layers.10.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
|
| 37 |
+
"model.layers.10.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 38 |
+
"model.layers.11.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 39 |
+
"model.layers.11.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 40 |
+
"model.layers.11.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 41 |
+
"model.layers.11.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 42 |
+
"model.layers.11.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 43 |
+
"model.layers.11.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 44 |
+
"model.layers.11.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 45 |
+
"model.layers.11.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 46 |
+
"model.layers.11.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
|
| 47 |
+
"model.layers.11.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 48 |
+
"model.layers.12.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 49 |
+
"model.layers.12.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 50 |
+
"model.layers.12.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 51 |
+
"model.layers.12.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 52 |
+
"model.layers.12.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 53 |
+
"model.layers.12.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 54 |
+
"model.layers.12.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 55 |
+
"model.layers.12.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 56 |
+
"model.layers.12.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
|
| 57 |
+
"model.layers.12.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 58 |
+
"model.layers.13.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 59 |
+
"model.layers.13.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 60 |
+
"model.layers.13.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 61 |
+
"model.layers.13.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 62 |
+
"model.layers.13.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 63 |
+
"model.layers.13.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 64 |
+
"model.layers.13.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 65 |
+
"model.layers.13.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 66 |
+
"model.layers.13.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
|
| 67 |
+
"model.layers.13.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 68 |
+
"model.layers.14.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 69 |
+
"model.layers.14.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 70 |
+
"model.layers.14.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 71 |
+
"model.layers.14.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 72 |
+
"model.layers.14.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 73 |
+
"model.layers.14.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 74 |
+
"model.layers.14.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 75 |
+
"model.layers.14.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 76 |
+
"model.layers.14.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
|
| 77 |
+
"model.layers.14.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 78 |
+
"model.layers.15.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 79 |
+
"model.layers.15.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 80 |
+
"model.layers.15.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 81 |
+
"model.layers.15.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 82 |
+
"model.layers.15.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 83 |
+
"model.layers.15.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 84 |
+
"model.layers.15.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 85 |
+
"model.layers.15.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 86 |
+
"model.layers.15.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
|
| 87 |
+
"model.layers.15.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 88 |
+
"model.layers.16.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 89 |
+
"model.layers.16.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 90 |
+
"model.layers.16.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 91 |
+
"model.layers.16.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 92 |
+
"model.layers.16.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 93 |
+
"model.layers.16.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 94 |
+
"model.layers.16.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 95 |
+
"model.layers.16.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 96 |
+
"model.layers.16.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
|
| 97 |
+
"model.layers.16.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 98 |
+
"model.layers.17.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 99 |
+
"model.layers.17.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 100 |
+
"model.layers.17.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 101 |
+
"model.layers.17.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 102 |
+
"model.layers.17.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 103 |
+
"model.layers.17.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 104 |
+
"model.layers.17.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 105 |
+
"model.layers.17.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 106 |
+
"model.layers.17.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
|
| 107 |
+
"model.layers.17.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 108 |
+
"model.layers.18.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 109 |
+
"model.layers.18.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 110 |
+
"model.layers.18.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 111 |
+
"model.layers.18.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 112 |
+
"model.layers.18.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 113 |
+
"model.layers.18.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 114 |
+
"model.layers.18.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 115 |
+
"model.layers.18.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 116 |
+
"model.layers.18.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
|
| 117 |
+
"model.layers.18.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 118 |
+
"model.layers.19.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 119 |
+
"model.layers.19.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 120 |
+
"model.layers.19.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 121 |
+
"model.layers.19.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 122 |
+
"model.layers.19.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 123 |
+
"model.layers.19.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 124 |
+
"model.layers.19.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 125 |
+
"model.layers.19.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 126 |
+
"model.layers.19.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
|
| 127 |
+
"model.layers.19.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 128 |
+
"model.layers.2.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 129 |
+
"model.layers.2.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 130 |
+
"model.layers.2.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 131 |
+
"model.layers.2.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 132 |
+
"model.layers.2.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 133 |
+
"model.layers.2.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 134 |
+
"model.layers.2.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 135 |
+
"model.layers.2.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 136 |
+
"model.layers.2.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
|
| 137 |
+
"model.layers.2.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 138 |
+
"model.layers.20.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 139 |
+
"model.layers.20.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 140 |
+
"model.layers.20.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 141 |
+
"model.layers.20.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 142 |
+
"model.layers.20.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 143 |
+
"model.layers.20.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 144 |
+
"model.layers.20.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 145 |
+
"model.layers.20.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 146 |
+
"model.layers.20.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
|
| 147 |
+
"model.layers.20.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 148 |
+
"model.layers.21.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 149 |
+
"model.layers.21.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 150 |
+
"model.layers.21.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 151 |
+
"model.layers.21.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 152 |
+
"model.layers.21.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 153 |
+
"model.layers.21.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 154 |
+
"model.layers.21.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 155 |
+
"model.layers.21.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 156 |
+
"model.layers.21.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
|
| 157 |
+
"model.layers.21.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 158 |
+
"model.layers.22.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 159 |
+
"model.layers.22.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 160 |
+
"model.layers.22.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 161 |
+
"model.layers.22.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 162 |
+
"model.layers.22.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 163 |
+
"model.layers.22.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 164 |
+
"model.layers.22.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 165 |
+
"model.layers.22.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 166 |
+
"model.layers.22.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
|
| 167 |
+
"model.layers.22.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 168 |
+
"model.layers.23.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 169 |
+
"model.layers.23.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 170 |
+
"model.layers.23.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 171 |
+
"model.layers.23.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 172 |
+
"model.layers.23.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 173 |
+
"model.layers.23.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 174 |
+
"model.layers.23.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 175 |
+
"model.layers.23.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 176 |
+
"model.layers.23.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
|
| 177 |
+
"model.layers.23.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 178 |
+
"model.layers.24.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 179 |
+
"model.layers.24.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 180 |
+
"model.layers.24.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 181 |
+
"model.layers.24.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 182 |
+
"model.layers.24.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 183 |
+
"model.layers.24.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 184 |
+
"model.layers.24.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 185 |
+
"model.layers.24.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 186 |
+
"model.layers.24.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
|
| 187 |
+
"model.layers.24.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 188 |
+
"model.layers.25.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 189 |
+
"model.layers.25.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 190 |
+
"model.layers.25.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 191 |
+
"model.layers.25.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 192 |
+
"model.layers.25.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 193 |
+
"model.layers.25.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 194 |
+
"model.layers.25.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 195 |
+
"model.layers.25.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 196 |
+
"model.layers.25.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
|
| 197 |
+
"model.layers.25.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 198 |
+
"model.layers.26.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 199 |
+
"model.layers.26.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 200 |
+
"model.layers.26.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 201 |
+
"model.layers.26.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 202 |
+
"model.layers.26.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 203 |
+
"model.layers.26.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 204 |
+
"model.layers.26.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 205 |
+
"model.layers.26.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 206 |
+
"model.layers.26.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
|
| 207 |
+
"model.layers.26.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 208 |
+
"model.layers.27.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 209 |
+
"model.layers.27.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 210 |
+
"model.layers.27.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 211 |
+
"model.layers.27.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 212 |
+
"model.layers.27.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 213 |
+
"model.layers.27.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 214 |
+
"model.layers.27.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 215 |
+
"model.layers.27.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 216 |
+
"model.layers.27.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
|
| 217 |
+
"model.layers.27.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 218 |
+
"model.layers.28.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 219 |
+
"model.layers.28.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 220 |
+
"model.layers.28.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 221 |
+
"model.layers.28.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 222 |
+
"model.layers.28.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 223 |
+
"model.layers.28.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 224 |
+
"model.layers.28.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 225 |
+
"model.layers.28.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 226 |
+
"model.layers.28.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
|
| 227 |
+
"model.layers.28.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 228 |
+
"model.layers.29.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 229 |
+
"model.layers.29.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 230 |
+
"model.layers.29.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 231 |
+
"model.layers.29.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 232 |
+
"model.layers.29.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
|
| 233 |
+
"model.layers.29.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 234 |
+
"model.layers.29.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 235 |
+
"model.layers.29.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 236 |
+
"model.layers.29.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
|
| 237 |
+
"model.layers.29.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
|
| 238 |
+
"model.layers.3.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 239 |
+
"model.layers.3.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 240 |
+
"model.layers.3.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 241 |
+
"model.layers.3.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 242 |
+
"model.layers.3.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 243 |
+
"model.layers.3.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 244 |
+
"model.layers.3.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 245 |
+
"model.layers.3.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 246 |
+
"model.layers.3.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
|
| 247 |
+
"model.layers.3.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 248 |
+
"model.layers.30.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 249 |
+
"model.layers.30.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 250 |
+
"model.layers.30.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 251 |
+
"model.layers.30.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 252 |
+
"model.layers.30.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 253 |
+
"model.layers.30.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 254 |
+
"model.layers.30.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 255 |
+
"model.layers.30.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 256 |
+
"model.layers.30.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
|
| 257 |
+
"model.layers.30.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 258 |
+
"model.layers.31.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 259 |
+
"model.layers.31.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 260 |
+
"model.layers.31.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 261 |
+
"model.layers.31.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 262 |
+
"model.layers.31.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 263 |
+
"model.layers.31.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 264 |
+
"model.layers.31.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 265 |
+
"model.layers.31.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 266 |
+
"model.layers.31.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
|
| 267 |
+
"model.layers.31.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 268 |
+
"model.layers.32.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 269 |
+
"model.layers.32.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 270 |
+
"model.layers.32.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 271 |
+
"model.layers.32.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 272 |
+
"model.layers.32.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 273 |
+
"model.layers.32.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 274 |
+
"model.layers.32.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 275 |
+
"model.layers.32.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 276 |
+
"model.layers.32.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
|
| 277 |
+
"model.layers.32.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 278 |
+
"model.layers.33.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 279 |
+
"model.layers.33.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 280 |
+
"model.layers.33.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 281 |
+
"model.layers.33.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 282 |
+
"model.layers.33.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 283 |
+
"model.layers.33.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 284 |
+
"model.layers.33.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 285 |
+
"model.layers.33.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 286 |
+
"model.layers.33.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
|
| 287 |
+
"model.layers.33.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 288 |
+
"model.layers.34.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 289 |
+
"model.layers.34.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 290 |
+
"model.layers.34.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 291 |
+
"model.layers.34.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 292 |
+
"model.layers.34.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 293 |
+
"model.layers.34.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 294 |
+
"model.layers.34.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 295 |
+
"model.layers.34.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 296 |
+
"model.layers.34.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
|
| 297 |
+
"model.layers.34.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 298 |
+
"model.layers.35.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 299 |
+
"model.layers.35.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 300 |
+
"model.layers.35.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 301 |
+
"model.layers.35.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 302 |
+
"model.layers.35.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 303 |
+
"model.layers.35.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 304 |
+
"model.layers.35.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 305 |
+
"model.layers.35.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 306 |
+
"model.layers.35.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
|
| 307 |
+
"model.layers.35.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 308 |
+
"model.layers.36.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 309 |
+
"model.layers.36.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 310 |
+
"model.layers.36.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 311 |
+
"model.layers.36.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 312 |
+
"model.layers.36.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 313 |
+
"model.layers.36.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 314 |
+
"model.layers.36.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 315 |
+
"model.layers.36.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 316 |
+
"model.layers.36.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
|
| 317 |
+
"model.layers.36.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 318 |
+
"model.layers.37.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 319 |
+
"model.layers.37.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 320 |
+
"model.layers.37.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 321 |
+
"model.layers.37.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 322 |
+
"model.layers.37.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 323 |
+
"model.layers.37.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 324 |
+
"model.layers.37.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 325 |
+
"model.layers.37.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 326 |
+
"model.layers.37.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
|
| 327 |
+
"model.layers.37.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 328 |
+
"model.layers.38.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 329 |
+
"model.layers.38.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 330 |
+
"model.layers.38.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 331 |
+
"model.layers.38.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 332 |
+
"model.layers.38.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 333 |
+
"model.layers.38.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 334 |
+
"model.layers.38.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 335 |
+
"model.layers.38.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 336 |
+
"model.layers.38.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
|
| 337 |
+
"model.layers.38.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 338 |
+
"model.layers.39.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 339 |
+
"model.layers.39.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 340 |
+
"model.layers.39.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 341 |
+
"model.layers.39.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 342 |
+
"model.layers.39.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
|
| 343 |
+
"model.layers.39.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 344 |
+
"model.layers.39.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 345 |
+
"model.layers.39.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 346 |
+
"model.layers.39.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
|
| 347 |
+
"model.layers.39.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
|
| 348 |
+
"model.layers.4.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 349 |
+
"model.layers.4.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 350 |
+
"model.layers.4.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 351 |
+
"model.layers.4.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 352 |
+
"model.layers.4.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 353 |
+
"model.layers.4.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 354 |
+
"model.layers.4.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 355 |
+
"model.layers.4.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 356 |
+
"model.layers.4.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
|
| 357 |
+
"model.layers.4.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 358 |
+
"model.layers.5.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 359 |
+
"model.layers.5.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 360 |
+
"model.layers.5.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 361 |
+
"model.layers.5.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 362 |
+
"model.layers.5.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 363 |
+
"model.layers.5.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 364 |
+
"model.layers.5.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 365 |
+
"model.layers.5.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 366 |
+
"model.layers.5.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
|
| 367 |
+
"model.layers.5.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 368 |
+
"model.layers.6.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 369 |
+
"model.layers.6.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 370 |
+
"model.layers.6.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 371 |
+
"model.layers.6.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 372 |
+
"model.layers.6.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 373 |
+
"model.layers.6.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 374 |
+
"model.layers.6.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 375 |
+
"model.layers.6.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 376 |
+
"model.layers.6.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
|
| 377 |
+
"model.layers.6.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 378 |
+
"model.layers.7.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 379 |
+
"model.layers.7.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 380 |
+
"model.layers.7.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 381 |
+
"model.layers.7.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 382 |
+
"model.layers.7.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 383 |
+
"model.layers.7.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 384 |
+
"model.layers.7.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 385 |
+
"model.layers.7.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 386 |
+
"model.layers.7.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
|
| 387 |
+
"model.layers.7.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 388 |
+
"model.layers.8.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 389 |
+
"model.layers.8.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 390 |
+
"model.layers.8.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 391 |
+
"model.layers.8.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 392 |
+
"model.layers.8.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 393 |
+
"model.layers.8.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 394 |
+
"model.layers.8.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 395 |
+
"model.layers.8.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 396 |
+
"model.layers.8.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
|
| 397 |
+
"model.layers.8.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 398 |
+
"model.layers.9.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 399 |
+
"model.layers.9.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 400 |
+
"model.layers.9.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 401 |
+
"model.layers.9.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 402 |
+
"model.layers.9.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
|
| 403 |
+
"model.layers.9.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 404 |
+
"model.layers.9.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 405 |
+
"model.layers.9.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 406 |
+
"model.layers.9.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
|
| 407 |
+
"model.layers.9.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
|
| 408 |
+
"model.norm.weight": "pytorch_model-00003-of-00003.bin"
|
| 409 |
+
}
|
| 410 |
+
}
|
special_tokens_map.json
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bos_token": {
|
| 3 |
+
"content": "<|startoftext|>",
|
| 4 |
+
"lstrip": false,
|
| 5 |
+
"normalized": false,
|
| 6 |
+
"rstrip": false,
|
| 7 |
+
"single_word": false
|
| 8 |
+
},
|
| 9 |
+
"eos_token": {
|
| 10 |
+
"content": "<|endoftext|>",
|
| 11 |
+
"lstrip": false,
|
| 12 |
+
"normalized": false,
|
| 13 |
+
"rstrip": false,
|
| 14 |
+
"single_word": false
|
| 15 |
+
},
|
| 16 |
+
"pad_token": {
|
| 17 |
+
"content": "<pad>",
|
| 18 |
+
"lstrip": false,
|
| 19 |
+
"normalized": false,
|
| 20 |
+
"rstrip": false,
|
| 21 |
+
"single_word": false
|
| 22 |
+
}
|
| 23 |
+
}
|
tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
tokenizer_config.json
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"clean_up_tokenization_spaces": true,
|
| 3 |
+
"model_max_length": 1000000000000000019884624838656,
|
| 4 |
+
"tokenizer_class": "PreTrainedTokenizerFast"
|
| 5 |
+
}
|