davidlvxin commited on
Commit
cae9942
1 Parent(s): e521bd8

Upload 9 files

Browse files
README.md ADDED
@@ -0,0 +1,79 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - en
4
+ - zh
5
+ library_name: transformers
6
+ tags:
7
+ - Long Context
8
+ - chatglm
9
+ - llama
10
+ datasets:
11
+ - THUDM/LongAlign-10k
12
+ - THUDM/LongBench
13
+ ---
14
+ # LongAlign-13B-64k-base
15
+
16
+ <p align="center">
17
+ 🤗 <a href="https://huggingface.co/datasets/THUDM/LongAlign-10k" target="_blank">[LongAlign Dataset] </a> • 💻 <a href="https://github.com/THUDM/LongAlign" target="_blank">[Github Repo]</a> • 📃 <a href="https://arxiv.org/" target="_blank">[LongAlign Paper]</a>
18
+ </p>
19
+
20
+ **LongAlign** is the first full recipe for LLM alignment on long context. We propose the **LongAlign-10k** dataset, containing 10,000 long instruction data of 8k-64k in length. We investigate on trianing strategies, namely **packing (with loss weighting) and sorted batching**, which are all implemented in our code. For real-world long context evaluation, we introduce **Chat-LongBench** that evaluate the instruction-following capability on queries of 10k-100k length.
21
+
22
+ ## All Models
23
+
24
+ We open-sourced the following list of models:
25
+
26
+ |Model|Huggingface Repo|Description|
27
+ |---|---|---|
28
+ |**LongAlign-6B-64k-base**| [🤗 Huggingface Repo](https://huggingface.co/THUDM/LongAlign-6B-64k-base) | **ChatGLM3-6B** with an extended 64k context window |
29
+ |**LongAlign-6B-64k**| [🤗 Huggingface Repo](https://huggingface.co/THUDM/LongAlign-6B-64k) | Chat model by LongAlign training on LongAlign-6B-64k-base|
30
+ |**LongAlign-7B-64k-base**| [🤗 Huggingface Repo](https://huggingface.co/THUDM/LongAlign-7B-64k-base) | **Llama-2-7B** with an extended 64k context window |
31
+ |**LongAlign-7B-64k**| [🤗 Huggingface Repo](https://huggingface.co/THUDM/LongAlign-7B-64k) | Chat model by LongAlign training on LongAlign-7B-64k-base|
32
+ |**LongAlign-13B-64k-base**| [🤗 Huggingface Repo](https://huggingface.co/THUDM/LongAlign-13B-64k-base) | **Llama-2-13B** with an extended 64k context window |
33
+ |**LongAlign-13B-64k**| [🤗 Huggingface Repo](https://huggingface.co/THUDM/LongAlign-13B-64k) | Chat model by LongAlign training on LongAlign-13B-64k-base|
34
+ |**ChatGLM3-6B-128k**| [🤗 Huggingface Repo](https://huggingface.co/THUDM/chatglm3-6b-128k) | **ChatGLM3-6B** with a 128k context window|
35
+
36
+ ![](assets/leaderboard.png)
37
+
38
+ ## Model usage
39
+ Chat prompt template for LongAlign-6B-64k:
40
+ ```text
41
+ [Round 1]
42
+
43
+ 问:Hi!
44
+
45
+ 答:Hello! What can I assist you today?
46
+
47
+ [Round 2]
48
+
49
+ 问:What should I do if I can't sleep at night?
50
+
51
+ 答:
52
+ ```
53
+ Chat prompt template for LongAlign-7B-64k and LongAlign-13B-64k:
54
+ ```text
55
+ [INST]Hi![/INST]Hello! What can I assist you today?
56
+
57
+ [INST]What should I do if I can't sleep at night?[/INST]
58
+ ```
59
+ ChatGLM3-6B-128k uses the same prompt template as [ChatGLM3-6B](https://huggingface.co/THUDM/chatglm3-6b).
60
+
61
+ A simple demo for deployment of the model:
62
+ ```python
63
+ from transformers import AutoTokenizer, AutoModelForCausalLM
64
+ import torch
65
+ tokenizer = AutoTokenizer.from_pretrained("THUDM/LongAlign-6B-64k", trust_remote_code=True)
66
+ model = AutoModelForCausalLM.from_pretrained("THUDM/LongAlign-6B-64k", torch_dtype=torch.bfloat16, trust_remote_code=True, device_map="auto")
67
+ model = model.eval()
68
+ query = open("assets/paper.txt").read() + "\n\nPlease summarize the paper."
69
+ response, history = model.chat(tokenizer, query, history=[], max_new_tokens=512, temperature=1)
70
+ print(response)
71
+ ```
72
+
73
+ ## Citation
74
+
75
+ If you find our work useful, please consider citing LongAlign:
76
+
77
+ ```
78
+
79
+ ```
assets/leaderboard.png ADDED
config.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "/mnt/vepfs/users/bys/fix_models/Llama-2-13b-64k-LongAlign-sort",
3
+ "architectures": [
4
+ "LlamaModel"
5
+ ],
6
+ "bos_token_id": 1,
7
+ "eos_token_id": 2,
8
+ "hidden_act": "silu",
9
+ "hidden_size": 5120,
10
+ "initializer_range": 0.02,
11
+ "intermediate_size": 13824,
12
+ "max_position_embeddings": 65536,
13
+ "max_sequence_length": 65536,
14
+ "model_type": "llama",
15
+ "num_attention_heads": 40,
16
+ "num_hidden_layers": 40,
17
+ "num_key_value_heads": 40,
18
+ "pad_token_id": 0,
19
+ "pretraining_tp": 1,
20
+ "rms_norm_eps": 1e-05,
21
+ "rope_scaling": null,
22
+ "rope_theta": 2000000.0,
23
+ "tie_word_embeddings": false,
24
+ "torch_dtype": "bfloat16",
25
+ "transformers_version": "4.33.0",
26
+ "use_cache": true,
27
+ "vocab_size": 32256
28
+ }
generation_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 1,
3
+ "eos_token_id": 2,
4
+ "max_length": 65536,
5
+ "pad_token_id": 0,
6
+ "temperature": 0.9,
7
+ "top_p": 0.6,
8
+ "transformers_version": "4.31.0"
9
+ }
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:192bbd9ab547444d92487acac6f0e102eecdd4ba0a9b81d11f2d7d57dcb673c5
3
+ size 26037067133
special_tokens_map.json ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "<s>",
4
+ "lstrip": false,
5
+ "normalized": false,
6
+ "rstrip": false,
7
+ "single_word": false
8
+ },
9
+ "eos_token": {
10
+ "content": "</s>",
11
+ "lstrip": false,
12
+ "normalized": false,
13
+ "rstrip": false,
14
+ "single_word": false
15
+ },
16
+ "pad_token": "<unk>",
17
+ "unk_token": {
18
+ "content": "<unk>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ }
24
+ }
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
3
+ size 499723
tokenizer_config.json ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": true,
3
+ "add_eos_token": false,
4
+ "bos_token": {
5
+ "__type": "AddedToken",
6
+ "content": "<s>",
7
+ "lstrip": false,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false
11
+ },
12
+ "clean_up_tokenization_spaces": false,
13
+ "eos_token": {
14
+ "__type": "AddedToken",
15
+ "content": "</s>",
16
+ "lstrip": false,
17
+ "normalized": false,
18
+ "rstrip": false,
19
+ "single_word": false
20
+ },
21
+ "legacy": false,
22
+ "model_max_length": 65536,
23
+ "pad_token": null,
24
+ "padding_side": "right",
25
+ "sp_model_kwargs": {},
26
+ "tokenizer_class": "LlamaTokenizer",
27
+ "unk_token": {
28
+ "__type": "AddedToken",
29
+ "content": "<unk>",
30
+ "lstrip": false,
31
+ "normalized": false,
32
+ "rstrip": false,
33
+ "single_word": false
34
+ }
35
+ }