TheBloke commited on
Commit
e6df405
1 Parent(s): 1069857

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +179 -46
README.md CHANGED
@@ -5,16 +5,9 @@ license: llama2
5
  model_creator: Xwin-LM
6
  model_name: Xwin-LM 13B V0.1
7
  model_type: llama
8
- prompt_template: 'Below is an instruction that describes a task. Write a response
9
- that appropriately completes the request.
10
-
11
-
12
- ### Instruction:
13
-
14
- {prompt}
15
-
16
-
17
- ### Response:
18
 
19
  '
20
  quantized_by: TheBloke
@@ -59,15 +52,10 @@ Multiple GPTQ parameter permutations are provided; see Provided Files below for
59
  <!-- repositories-available end -->
60
 
61
  <!-- prompt-template start -->
62
- ## Prompt template: Alpaca
63
 
64
  ```
65
- Below is an instruction that describes a task. Write a response that appropriately completes the request.
66
-
67
- ### Instruction:
68
- {prompt}
69
-
70
- ### Response:
71
 
72
  ```
73
 
@@ -75,7 +63,7 @@ Below is an instruction that describes a task. Write a response that appropriate
75
 
76
 
77
  <!-- README_GPTQ.md-provided-files start -->
78
- ## Provided files and GPTQ parameters
79
 
80
  Multiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements.
81
 
@@ -90,8 +78,8 @@ All recent GPTQ files are made with AutoGPTQ, and all files in non-main branches
90
  - GS: GPTQ group size. Higher numbers use less VRAM, but have lower quantisation accuracy. "None" is the lowest possible value.
91
  - Act Order: True or False. Also known as `desc_act`. True results in better quantisation accuracy. Some GPTQ clients have had issues with models that use Act Order plus Group Size, but this is generally resolved now.
92
  - Damp %: A GPTQ parameter that affects how samples are processed for quantisation. 0.01 is default, but 0.1 results in slightly better accuracy.
93
- - GPTQ dataset: The dataset used for quantisation. Using a dataset more appropriate to the model's training can improve quantisation accuracy. Note that the GPTQ dataset is not the same as the dataset used to train the model - please refer to the original model repo for details of the training dataset(s).
94
- - Sequence Length: The length of the dataset sequences used for quantisation. Ideally this is the same as the model sequence length. For some very long sequence models (16+K), a lower sequence length may have to be used. Note that a lower sequence length does not limit the sequence length of the quantised model. It only impacts the quantisation accuracy on longer inference sequences.
95
  - ExLlama Compatibility: Whether this file can be loaded with ExLlama, which currently only supports Llama models in 4-bit.
96
 
97
  </details>
@@ -107,14 +95,71 @@ All recent GPTQ files are made with AutoGPTQ, and all files in non-main branches
107
  <!-- README_GPTQ.md-provided-files end -->
108
 
109
  <!-- README_GPTQ.md-download-from-branches start -->
110
- ## How to download from branches
 
 
 
 
 
 
111
 
112
- - In text-generation-webui, you can add `:branch` to the end of the download name, eg `TheBloke/Xwin-LM-13B-V0.1-GPTQ:main`
113
- - With Git, you can clone a branch with:
 
 
 
 
 
 
 
 
 
 
 
114
  ```
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
115
  git clone --single-branch --branch main https://huggingface.co/TheBloke/Xwin-LM-13B-V0.1-GPTQ
116
  ```
117
- - In Python Transformers code, the branch is the `revision` parameter; see below.
 
 
118
  <!-- README_GPTQ.md-download-from-branches end -->
119
  <!-- README_GPTQ.md-text-generation-webui start -->
120
  ## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
@@ -142,10 +187,10 @@ It is strongly recommended to use the text-generation-webui one-click-installers
142
 
143
  ### Install the necessary packages
144
 
145
- Requires: Transformers 4.32.0 or later, Optimum 1.12.0 or later, and AutoGPTQ 0.4.2 or later.
146
 
147
  ```shell
148
- pip3 install transformers>=4.32.0 optimum>=1.12.0
149
  pip3 install auto-gptq --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/ # Use cu117 if on CUDA 11.7
150
  ```
151
 
@@ -155,17 +200,10 @@ If you have problems installing AutoGPTQ using the pre-built wheels, install it
155
  pip3 uninstall -y auto-gptq
156
  git clone https://github.com/PanQiWei/AutoGPTQ
157
  cd AutoGPTQ
 
158
  pip3 install .
159
  ```
160
 
161
- ### For CodeLlama models only: you must use Transformers 4.33.0 or later.
162
-
163
- If 4.33.0 is not yet released when you read this, you will need to install Transformers from source:
164
- ```shell
165
- pip3 uninstall -y transformers
166
- pip3 install git+https://github.com/huggingface/transformers.git
167
- ```
168
-
169
  ### You can then use the following code
170
 
171
  ```python
@@ -182,13 +220,7 @@ model = AutoModelForCausalLM.from_pretrained(model_name_or_path,
182
  tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
183
 
184
  prompt = "Tell me about AI"
185
- prompt_template=f'''Below is an instruction that describes a task. Write a response that appropriately completes the request.
186
-
187
- ### Instruction:
188
- {prompt}
189
-
190
- ### Response:
191
-
192
  '''
193
 
194
  print("\n\n*** Generate:")
@@ -268,6 +300,9 @@ Xwin-LM: Powerful, Stable, and Reproducible LLM Alignment
268
  </h3>
269
 
270
  <p align="center">
 
 
 
271
  <a href="https://huggingface.co/Xwin-LM">
272
  <img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Models-blue">
273
  </a>
@@ -277,13 +312,14 @@ Xwin-LM: Powerful, Stable, and Reproducible LLM Alignment
277
 
278
  **Step up your LLM alignment with Xwin-LM!**
279
 
280
- Xwin-LM aims to develop and open-source alignment technologies for large language models, including supervised fine-tuning (SFT), reward models, reject sampling, reinforcement learning, etc. Our first release, built-upon on the Llama2 base models, ranked **TOP-1** on [AlpacaEval](https://tatsu-lab.github.io/alpaca_eval/). Notably, it's **the first to surpass GPT-4** on this benchmark. The project will be continuously updated.
281
 
282
  ## News
283
 
284
- - :boom: [Sep, 2023] We released [Xwin-LM-70B-V0.1](https://huggingface.co/Xwin-LM/Xwin-LM-70B-V0.1), which has achieved a win-rate against Davinci-003 of **95.57%** on [AlpacaEval](https://tatsu-lab.github.io/alpaca_eval/) benchmark, ranking as **TOP-1** on AlpacaEval. **It was the FIRST model surpassing GPT-4** on [AlpacaEval](https://tatsu-lab.github.io/alpaca_eval/). Also note its winrate v.s. GPT-4 is **60.61**.
285
- - :boom: [Sep, 2023] We released [Xwin-LM-13B-V0.1](https://huggingface.co/Xwin-LM/Xwin-LM-13B-V0.1), which has achieved **91.76%** win-rate on [AlpacaEval](https://tatsu-lab.github.io/alpaca_eval/), ranking as **top-1** among all 13B models.
286
- - :boom: [Sep, 2023] We released [Xwin-LM-7B-V0.1](https://huggingface.co/Xwin-LM/Xwin-LM-7B-V0.1), which has achieved **87.82%** win-rate on [AlpacaEval](https://tatsu-lab.github.io/alpaca_eval/), ranking as **top-1** among all 7B models.
 
287
 
288
 
289
  ## Model Card
@@ -311,4 +347,101 @@ The table below displays the performance of Xwin-LM on [AlpacaEval](https://tats
311
  | **Xwin-LM-7B-V0.1** | **87.35** | **76.40** | **47.57** |
312
  | Llama-2-13B-Chat | 81.09 | 64.22 | 30.92 |
313
 
314
- ##
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5
  model_creator: Xwin-LM
6
  model_name: Xwin-LM 13B V0.1
7
  model_type: llama
8
+ prompt_template: 'A chat between a curious user and an artificial intelligence assistant.
9
+ The assistant gives helpful, detailed, and polite answers to the user''s questions.
10
+ USER: {prompt} ASSISTANT:
 
 
 
 
 
 
 
11
 
12
  '
13
  quantized_by: TheBloke
 
52
  <!-- repositories-available end -->
53
 
54
  <!-- prompt-template start -->
55
+ ## Prompt template: Vicuna
56
 
57
  ```
58
+ A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. USER: {prompt} ASSISTANT:
 
 
 
 
 
59
 
60
  ```
61
 
 
63
 
64
 
65
  <!-- README_GPTQ.md-provided-files start -->
66
+ ## Provided files, and GPTQ parameters
67
 
68
  Multiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements.
69
 
 
78
  - GS: GPTQ group size. Higher numbers use less VRAM, but have lower quantisation accuracy. "None" is the lowest possible value.
79
  - Act Order: True or False. Also known as `desc_act`. True results in better quantisation accuracy. Some GPTQ clients have had issues with models that use Act Order plus Group Size, but this is generally resolved now.
80
  - Damp %: A GPTQ parameter that affects how samples are processed for quantisation. 0.01 is default, but 0.1 results in slightly better accuracy.
81
+ - GPTQ dataset: The calibration dataset used during quantisation. Using a dataset more appropriate to the model's training can improve quantisation accuracy. Note that the GPTQ calibration dataset is not the same as the dataset used to train the model - please refer to the original model repo for details of the training dataset(s).
82
+ - Sequence Length: The length of the dataset sequences used for quantisation. Ideally this is the same as the model sequence length. For some very long sequence models (16+K), a lower sequence length may have to be used. Note that a lower sequence length does not limit the sequence length of the quantised model. It only impacts the quantisation accuracy on longer inference sequences.
83
  - ExLlama Compatibility: Whether this file can be loaded with ExLlama, which currently only supports Llama models in 4-bit.
84
 
85
  </details>
 
95
  <!-- README_GPTQ.md-provided-files end -->
96
 
97
  <!-- README_GPTQ.md-download-from-branches start -->
98
+ ## How to download, including from branches
99
+
100
+ ### In text-generation-webui
101
+
102
+ To download from the `main` branch, enter `TheBloke/Xwin-LM-13B-V0.1-GPTQ` in the "Download model" box.
103
+
104
+ To download from another branch, add `:branchname` to the end of the download name, eg `TheBloke/Xwin-LM-13B-V0.1-GPTQ:main`
105
 
106
+ ### From the command line
107
+
108
+ I recommend using the `huggingface-hub` Python library:
109
+
110
+ ```shell
111
+ pip3 install huggingface-hub
112
+ ```
113
+
114
+ To download the `main` branch to a folder called `Xwin-LM-13B-V0.1-GPTQ`:
115
+
116
+ ```shell
117
+ mkdir Xwin-LM-13B-V0.1-GPTQ
118
+ huggingface-cli download TheBloke/Xwin-LM-13B-V0.1-GPTQ --local-dir Xwin-LM-13B-V0.1-GPTQ --local-dir-use-symlinks False
119
  ```
120
+
121
+ To download from a different branch, add the `--revision` parameter:
122
+
123
+ ```shell
124
+ mkdir Xwin-LM-13B-V0.1-GPTQ
125
+ huggingface-cli download TheBloke/Xwin-LM-13B-V0.1-GPTQ --revision main --local-dir Xwin-LM-13B-V0.1-GPTQ --local-dir-use-symlinks False
126
+ ```
127
+
128
+ <details>
129
+ <summary>More advanced huggingface-cli download usage</summary>
130
+
131
+ If you remove the `--local-dir-use-symlinks False` parameter, the files will instead be stored in the central Huggingface cache directory (default location on Linux is: `~/.cache/huggingface`), and symlinks will be added to the specified `--local-dir`, pointing to their real location in the cache. This allows for interrupted downloads to be resumed, and allows you to quickly clone the repo to multiple places on disk without triggering a download again. The downside, and the reason why I don't list that as the default option, is that the files are then hidden away in a cache folder and it's harder to know where your disk space is being used, and to clear it up if/when you want to remove a download model.
132
+
133
+ The cache location can be changed with the `HF_HOME` environment variable, and/or the `--cache-dir` parameter to `huggingface-cli`.
134
+
135
+ For more documentation on downloading with `huggingface-cli`, please see: [HF -> Hub Python Library -> Download files -> Download from the CLI](https://huggingface.co/docs/huggingface_hub/guides/download#download-from-the-cli).
136
+
137
+ To accelerate downloads on fast connections (1Gbit/s or higher), install `hf_transfer`:
138
+
139
+ ```shell
140
+ pip3 install hf_transfer
141
+ ```
142
+
143
+ And set environment variable `HF_HUB_ENABLE_HF_TRANSFER` to `1`:
144
+
145
+ ```shell
146
+ mkdir Xwin-LM-13B-V0.1-GPTQ
147
+ HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download TheBloke/Xwin-LM-13B-V0.1-GPTQ --local-dir Xwin-LM-13B-V0.1-GPTQ --local-dir-use-symlinks False
148
+ ```
149
+
150
+ Windows Command Line users: You can set the environment variable by running `set HF_HUB_ENABLE_HF_TRANSFER=1` before the download command.
151
+ </details>
152
+
153
+ ### With `git` (**not** recommended)
154
+
155
+ To clone a specific branch with `git`, use a command like this:
156
+
157
+ ```shell
158
  git clone --single-branch --branch main https://huggingface.co/TheBloke/Xwin-LM-13B-V0.1-GPTQ
159
  ```
160
+
161
+ Note that using Git with HF repos is strongly discouraged. It will be much slower than using `huggingface-hub`, and will use twice as much disk space as it has to store the model files twice (it stores every byte both in the intended target folder, and again in the `.git` folder as a blob.)
162
+
163
  <!-- README_GPTQ.md-download-from-branches end -->
164
  <!-- README_GPTQ.md-text-generation-webui start -->
165
  ## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
 
187
 
188
  ### Install the necessary packages
189
 
190
+ Requires: Transformers 4.33.0 or later, Optimum 1.12.0 or later, and AutoGPTQ 0.4.2 or later.
191
 
192
  ```shell
193
+ pip3 install transformers optimum
194
  pip3 install auto-gptq --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/ # Use cu117 if on CUDA 11.7
195
  ```
196
 
 
200
  pip3 uninstall -y auto-gptq
201
  git clone https://github.com/PanQiWei/AutoGPTQ
202
  cd AutoGPTQ
203
+ git checkout v0.4.2
204
  pip3 install .
205
  ```
206
 
 
 
 
 
 
 
 
 
207
  ### You can then use the following code
208
 
209
  ```python
 
220
  tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
221
 
222
  prompt = "Tell me about AI"
223
+ prompt_template=f'''A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. USER: {prompt} ASSISTANT:
 
 
 
 
 
 
224
  '''
225
 
226
  print("\n\n*** Generate:")
 
300
  </h3>
301
 
302
  <p align="center">
303
+ <a href="https://github.com/Xwin-LM/Xwin-LM">
304
+ <img src="https://img.shields.io/badge/GitHub-yellow.svg?style=social&logo=github">
305
+ </a>
306
  <a href="https://huggingface.co/Xwin-LM">
307
  <img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Models-blue">
308
  </a>
 
312
 
313
  **Step up your LLM alignment with Xwin-LM!**
314
 
315
+ Xwin-LM aims to develop and open-source alignment technologies for large language models, including supervised fine-tuning (SFT), reward models (RM), reject sampling, reinforcement learning from human feedback (RLHF), etc. Our first release, built-upon on the Llama2 base models, ranked **TOP-1** on [AlpacaEval](https://tatsu-lab.github.io/alpaca_eval/). Notably, it's **the first to surpass GPT-4** on this benchmark. The project will be continuously updated.
316
 
317
  ## News
318
 
319
+ - 💥 [Sep, 2023] We released [Xwin-LM-70B-V0.1](https://huggingface.co/Xwin-LM/Xwin-LM-70B-V0.1), which has achieved a win-rate against Davinci-003 of **95.57%** on [AlpacaEval](https://tatsu-lab.github.io/alpaca_eval/) benchmark, ranking as **TOP-1** on AlpacaEval. **It was the FIRST model surpassing GPT-4** on [AlpacaEval](https://tatsu-lab.github.io/alpaca_eval/). Also note its winrate v.s. GPT-4 is **60.61**.
320
+ - 🔍 [Sep, 2023] RLHF plays crucial role in the strong performance of Xwin-LM-V0.1 release!
321
+ - 💥 [Sep, 2023] We released [Xwin-LM-13B-V0.1](https://huggingface.co/Xwin-LM/Xwin-LM-13B-V0.1), which has achieved **91.76%** win-rate on [AlpacaEval](https://tatsu-lab.github.io/alpaca_eval/), ranking as **top-1** among all 13B models.
322
+ - 💥 [Sep, 2023] We released [Xwin-LM-7B-V0.1](https://huggingface.co/Xwin-LM/Xwin-LM-7B-V0.1), which has achieved **87.82%** win-rate on [AlpacaEval](https://tatsu-lab.github.io/alpaca_eval/), ranking as **top-1** among all 7B models.
323
 
324
 
325
  ## Model Card
 
347
  | **Xwin-LM-7B-V0.1** | **87.35** | **76.40** | **47.57** |
348
  | Llama-2-13B-Chat | 81.09 | 64.22 | 30.92 |
349
 
350
+ ### Case study on [AlpacaEval](https://tatsu-lab.github.io/alpaca_eval/)
351
+
352
+ | Prompts | GTP4 response | Xwin-LM response |
353
+ |----------------|------------|--------|
354
+ |I've put a plate on top of an apple, and take the plate from kitchen to dinning room. Where is the apple now|The apple is now in the dining room, as it was underneath the plate when you moved the plate from the kitchen to the dining room.|The apple is still in the kitchen, as you have only moved the plate. The apple is under the plate, so if you want to bring the apple to the dining room, you should either move the apple directly or put the apple on the plate and then move the plate to the dining room.|
355
+ | Answer the following question.<br> How do you say "good evening" in French. | In French, "good evening" is said as "bonsoir." | In French, "good evening" is translated to "bonsoir." This greeting is used in the evening and nighttime hours, typically after 6 p.m. It is a combination of "bon," which means "good," and "soir," which means "evening." To pronounce it, say "bone-swahr." |
356
+
357
+
358
+ ### Xwin-LM performance on NLP foundation tasks.
359
+
360
+ The following table provides a comparison of Xwin-LMs with other LLMs on NLP foundation tasks in [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).
361
+
362
+ | Model | MMLU 5-shot | ARC 25-shot | TruthfulQA 0-shot | HellaSwag 10-shot | Average |
363
+ |------------------|-------------|-------------|-------------------|-------------------|------------|
364
+ | Text-davinci-003 | 56.9 | **85.2** | 59.3 | 82.2 | 70.9 |
365
+ |Vicuna-13b 1.1 | 51.3 | 53.0 | 51.8 | 80.1 | 59.1 |
366
+ |Guanaco 30B | 57.6 | 63.7 | 50.7 | 85.1 | 64.3 |
367
+ | WizardLM-7B 1.0 | 42.7 | 51.6 | 44.7 | 77.7 | 54.2 |
368
+ | WizardLM-13B 1.0 | 52.3 | 57.2 | 50.5 | 81.0 | 60.2 |
369
+ | WizardLM-30B 1.0 | 58.8 | 62.5 | 52.4 | 83.3 | 64.2|
370
+ | Llama-2-7B-Chat | 48.3 | 52.9 | 45.6 | 78.6 | 56.4 |
371
+ | Llama-2-13B-Chat | 54.6 | 59.0 | 44.1 | 81.9 | 59.9 |
372
+ | Llama-2-70B-Chat | 63.9 | 64.6 | 52.8 | 85.9 | 66.8 |
373
+ | **Xwin-LM-7B-V0.1** | 49.7 | 56.2 | 48.1 | 79.5 | 58.4 |
374
+ | **Xwin-LM-13B-V0.1** | 56.6 | 62.4 | 45.5 | 83.0 | 61.9 |
375
+ | **Xwin-LM-70B-V0.1** | **69.6** | 70.5 | **60.1** | **87.1** | **71.8** |
376
+
377
+
378
+ ## Inference
379
+
380
+ ### Conversation templates
381
+ To obtain desired results, please strictly follow the conversation templates when utilizing our model for inference. Our model adopts the prompt format established by [Vicuna](https://github.com/lm-sys/FastChat) and is equipped to support **multi-turn** conversations.
382
+ ```
383
+ A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. USER: Hi! ASSISTANT: Hello.</s>USER: Who are you? ASSISTANT: I am Xwin-LM.</s>......
384
+ ```
385
+
386
+ ### HuggingFace Example
387
+
388
+ ```python
389
+ from transformers import AutoTokenizer, AutoModelForCausalLM
390
+
391
+ model = AutoModelForCausalLM.from_pretrained("Xwin-LM/Xwin-LM-7B-V0.1")
392
+ tokenizer = AutoTokenizer.from_pretrained("Xwin-LM/Xwin-LM-7B-V0.1")
393
+ (
394
+ prompt := "A chat between a curious user and an artificial intelligence assistant. "
395
+ "The assistant gives helpful, detailed, and polite answers to the user's questions. "
396
+ "USER: Hello, can you help me? "
397
+ "ASSISTANT:"
398
+ )
399
+ inputs = tokenizer(prompt, return_tensors="pt")
400
+ samples = model.generate(**inputs, max_new_tokens=4096, temperature=0.7)
401
+ output = tokenizer.decode(samples[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True)
402
+ print(output)
403
+ # Of course! I'm here to help. Please feel free to ask your question or describe the issue you're having, and I'll do my best to assist you.
404
+ ```
405
+
406
+
407
+ ### vllm Example
408
+ Because Xwin-LM is based on Llama2, it also offers support for rapid inference using [vllm](https://github.com/vllm-project/vllm). Please refer to [vllm](https://github.com/vllm-project/vllm) for detailed installation instructions.
409
+ ```python
410
+ from vllm import LLM, SamplingParams
411
+ (
412
+ prompt := "A chat between a curious user and an artificial intelligence assistant. "
413
+ "The assistant gives helpful, detailed, and polite answers to the user's questions. "
414
+ "USER: Hello, can you help me? "
415
+ "ASSISTANT:"
416
+ )
417
+ sampling_params = SamplingParams(temperature=0.7, max_tokens=4096)
418
+ llm = LLM(model="Xwin-LM/Xwin-LM-7B-V0.1")
419
+ outputs = llm.generate([prompt,], sampling_params)
420
+
421
+ for output in outputs:
422
+ prompt = output.prompt
423
+ generated_text = output.outputs[0].text
424
+ print(generated_text)
425
+ ```
426
+
427
+ ## TODO
428
+
429
+ - [ ] Release the source code
430
+ - [ ] Release more capabilities, such as math, reasoning, and etc.
431
+
432
+ ## Citation
433
+ Please consider citing our work if you use the data or code in this repo.
434
+ ```
435
+ @software{xwin-lm,
436
+ title = {Xwin-LM},
437
+ author = {Xwin-LM Team},
438
+ url = {https://github.com/Xwin-LM/Xwin-LM},
439
+ version = {pre-release},
440
+ year = {2023},
441
+ month = {9},
442
+ }
443
+ ```
444
+
445
+ ## Acknowledgements
446
+
447
+ Thanks to [Llama 2](https://ai.meta.com/llama/), [FastChat](https://github.com/lm-sys/FastChat), [AlpacaFarm](https://github.com/tatsu-lab/alpaca_farm), and [vllm](https://github.com/vllm-project/vllm).