Text Generation
Transformers
Safetensors
English
llama
legal
text-generation-inference
4-bit precision
awq
TheBloke commited on
Commit
50f7d38
1 Parent(s): b8e0142

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +445 -0
README.md ADDED
@@ -0,0 +1,445 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: AdaptLLM/law-LLM-13B
3
+ datasets:
4
+ - Open-Orca/OpenOrca
5
+ - GAIR/lima
6
+ - WizardLM/WizardLM_evol_instruct_V2_196k
7
+ - EleutherAI/pile
8
+ inference: false
9
+ language:
10
+ - en
11
+ license: other
12
+ metrics:
13
+ - accuracy
14
+ model_creator: AdaptLLM
15
+ model_name: Law LLM 13B
16
+ model_type: llama
17
+ pipeline_tag: text-generation
18
+ prompt_template: '[INST] <<SYS>>
19
+
20
+ {system_message}
21
+
22
+ <</SYS>>
23
+
24
+ {prompt} [/INST]
25
+
26
+ '
27
+ quantized_by: TheBloke
28
+ tags:
29
+ - legal
30
+ ---
31
+ <!-- markdownlint-disable MD041 -->
32
+
33
+ <!-- header start -->
34
+ <!-- 200823 -->
35
+ <div style="width: auto; margin-left: auto; margin-right: auto">
36
+ <img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
37
+ </div>
38
+ <div style="display: flex; justify-content: space-between; width: 100%;">
39
+ <div style="display: flex; flex-direction: column; align-items: flex-start;">
40
+ <p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
41
+ </div>
42
+ <div style="display: flex; flex-direction: column; align-items: flex-end;">
43
+ <p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
44
+ </div>
45
+ </div>
46
+ <div style="text-align:center; margin-top: 0em; margin-bottom: 0em"><p style="margin-top: 0.25em; margin-bottom: 0em;">TheBloke's LLM work is generously supported by a grant from <a href="https://a16z.com">andreessen horowitz (a16z)</a></p></div>
47
+ <hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
48
+ <!-- header end -->
49
+
50
+ # Law LLM 13B - AWQ
51
+ - Model creator: [AdaptLLM](https://huggingface.co/AdaptLLM)
52
+ - Original model: [Law LLM 13B](https://huggingface.co/AdaptLLM/law-LLM-13B)
53
+
54
+ <!-- description start -->
55
+ ## Description
56
+
57
+ This repo contains AWQ model files for [AdaptLLM's Law LLM 13B](https://huggingface.co/AdaptLLM/law-LLM-13B).
58
+
59
+ These files were quantised using hardware kindly provided by [Massed Compute](https://massedcompute.com/).
60
+
61
+
62
+ ### About AWQ
63
+
64
+ AWQ is an efficient, accurate and blazing-fast low-bit weight quantization method, currently supporting 4-bit quantization. Compared to GPTQ, it offers faster Transformers-based inference with equivalent or better quality compared to the most commonly used GPTQ settings.
65
+
66
+ AWQ models are currently supported on Linux and Windows, with NVidia GPUs only. macOS users: please use GGUF models instead.
67
+
68
+ It is supported by:
69
+
70
+ - [Text Generation Webui](https://github.com/oobabooga/text-generation-webui) - using Loader: AutoAWQ
71
+ - [vLLM](https://github.com/vllm-project/vllm) - version 0.2.2 or later for support for all model types.
72
+ - [Hugging Face Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference)
73
+ - [Transformers](https://huggingface.co/docs/transformers) version 4.35.0 and later, from any code or client that supports Transformers
74
+ - [AutoAWQ](https://github.com/casper-hansen/AutoAWQ) - for use from Python code
75
+
76
+ <!-- description end -->
77
+ <!-- repositories-available start -->
78
+ ## Repositories available
79
+
80
+ * [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/law-LLM-13B-AWQ)
81
+ * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/law-LLM-13B-GPTQ)
82
+ * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/law-LLM-13B-GGUF)
83
+ * [AdaptLLM's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/AdaptLLM/law-LLM-13B)
84
+ <!-- repositories-available end -->
85
+
86
+ <!-- prompt-template start -->
87
+ ## Prompt template: Llama-2-Chat
88
+
89
+ ```
90
+ [INST] <<SYS>>
91
+ {system_message}
92
+ <</SYS>>
93
+ {prompt} [/INST]
94
+
95
+ ```
96
+
97
+ <!-- prompt-template end -->
98
+
99
+
100
+ <!-- README_AWQ.md-provided-files start -->
101
+ ## Provided files, and AWQ parameters
102
+
103
+ I currently release 128g GEMM models only. The addition of group_size 32 models, and GEMV kernel models, is being actively considered.
104
+
105
+ Models are released as sharded safetensors files.
106
+
107
+ | Branch | Bits | GS | AWQ Dataset | Seq Len | Size |
108
+ | ------ | ---- | -- | ----------- | ------- | ---- |
109
+ | [main](https://huggingface.co/TheBloke/law-LLM-13B-AWQ/tree/main) | 4 | 128 | [VMware Open Instruct](https://huggingface.co/datasets/VMware/open-instruct/viewer/) | 2048 | 7.25 GB
110
+
111
+ <!-- README_AWQ.md-provided-files end -->
112
+
113
+ <!-- README_AWQ.md-text-generation-webui start -->
114
+ ## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui)
115
+
116
+ Please make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui).
117
+
118
+ It is strongly recommended to use the text-generation-webui one-click-installers unless you're sure you know how to make a manual install.
119
+
120
+ 1. Click the **Model tab**.
121
+ 2. Under **Download custom model or LoRA**, enter `TheBloke/law-LLM-13B-AWQ`.
122
+ 3. Click **Download**.
123
+ 4. The model will start downloading. Once it's finished it will say "Done".
124
+ 5. In the top left, click the refresh icon next to **Model**.
125
+ 6. In the **Model** dropdown, choose the model you just downloaded: `law-LLM-13B-AWQ`
126
+ 7. Select **Loader: AutoAWQ**.
127
+ 8. Click Load, and the model will load and is now ready for use.
128
+ 9. If you want any custom settings, set them and then click **Save settings for this model** followed by **Reload the Model** in the top right.
129
+ 10. Once you're ready, click the **Text Generation** tab and enter a prompt to get started!
130
+ <!-- README_AWQ.md-text-generation-webui end -->
131
+
132
+ <!-- README_AWQ.md-use-from-vllm start -->
133
+ ## Multi-user inference server: vLLM
134
+
135
+ Documentation on installing and using vLLM [can be found here](https://vllm.readthedocs.io/en/latest/).
136
+
137
+ - Please ensure you are using vLLM version 0.2 or later.
138
+ - When using vLLM as a server, pass the `--quantization awq` parameter.
139
+
140
+ For example:
141
+
142
+ ```shell
143
+ python3 -m vllm.entrypoints.api_server --model TheBloke/law-LLM-13B-AWQ --quantization awq --dtype auto
144
+ ```
145
+
146
+ - When using vLLM from Python code, again set `quantization=awq`.
147
+
148
+ For example:
149
+
150
+ ```python
151
+ from vllm import LLM, SamplingParams
152
+
153
+ prompts = [
154
+ "Tell me about AI",
155
+ "Write a story about llamas",
156
+ "What is 291 - 150?",
157
+ "How much wood would a woodchuck chuck if a woodchuck could chuck wood?",
158
+ ]
159
+ prompt_template=f'''[INST] <<SYS>>
160
+ {system_message}
161
+ <</SYS>>
162
+ {prompt} [/INST]
163
+ '''
164
+
165
+ prompts = [prompt_template.format(prompt=prompt) for prompt in prompts]
166
+
167
+ sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
168
+
169
+ llm = LLM(model="TheBloke/law-LLM-13B-AWQ", quantization="awq", dtype="auto")
170
+
171
+ outputs = llm.generate(prompts, sampling_params)
172
+
173
+ # Print the outputs.
174
+ for output in outputs:
175
+ prompt = output.prompt
176
+ generated_text = output.outputs[0].text
177
+ print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
178
+ ```
179
+ <!-- README_AWQ.md-use-from-vllm start -->
180
+
181
+ <!-- README_AWQ.md-use-from-tgi start -->
182
+ ## Multi-user inference server: Hugging Face Text Generation Inference (TGI)
183
+
184
+ Use TGI version 1.1.0 or later. The official Docker container is: `ghcr.io/huggingface/text-generation-inference:1.1.0`
185
+
186
+ Example Docker parameters:
187
+
188
+ ```shell
189
+ --model-id TheBloke/law-LLM-13B-AWQ --port 3000 --quantize awq --max-input-length 3696 --max-total-tokens 4096 --max-batch-prefill-tokens 4096
190
+ ```
191
+
192
+ Example Python code for interfacing with TGI (requires [huggingface-hub](https://github.com/huggingface/huggingface_hub) 0.17.0 or later):
193
+
194
+ ```shell
195
+ pip3 install huggingface-hub
196
+ ```
197
+
198
+ ```python
199
+ from huggingface_hub import InferenceClient
200
+
201
+ endpoint_url = "https://your-endpoint-url-here"
202
+
203
+ prompt = "Tell me about AI"
204
+ prompt_template=f'''[INST] <<SYS>>
205
+ {system_message}
206
+ <</SYS>>
207
+ {prompt} [/INST]
208
+ '''
209
+
210
+ client = InferenceClient(endpoint_url)
211
+ response = client.text_generation(prompt,
212
+ max_new_tokens=128,
213
+ do_sample=True,
214
+ temperature=0.7,
215
+ top_p=0.95,
216
+ top_k=40,
217
+ repetition_penalty=1.1)
218
+
219
+ print(f"Model output: ", response)
220
+ ```
221
+ <!-- README_AWQ.md-use-from-tgi end -->
222
+
223
+ <!-- README_AWQ.md-use-from-python start -->
224
+ ## Inference from Python code using Transformers
225
+
226
+ ### Install the necessary packages
227
+
228
+ - Requires: [Transformers](https://huggingface.co/docs/transformers) 4.35.0 or later.
229
+ - Requires: [AutoAWQ](https://github.com/casper-hansen/AutoAWQ) 0.1.6 or later.
230
+
231
+ ```shell
232
+ pip3 install --upgrade "autoawq>=0.1.6" "transformers>=4.35.0"
233
+ ```
234
+
235
+ Note that if you are using PyTorch 2.0.1, the above AutoAWQ command will automatically upgrade you to PyTorch 2.1.0.
236
+
237
+ If you are using CUDA 11.8 and wish to continue using PyTorch 2.0.1, instead run this command:
238
+
239
+ ```shell
240
+ pip3 install https://github.com/casper-hansen/AutoAWQ/releases/download/v0.1.6/autoawq-0.1.6+cu118-cp310-cp310-linux_x86_64.whl
241
+ ```
242
+
243
+ If you have problems installing [AutoAWQ](https://github.com/casper-hansen/AutoAWQ) using the pre-built wheels, install it from source instead:
244
+
245
+ ```shell
246
+ pip3 uninstall -y autoawq
247
+ git clone https://github.com/casper-hansen/AutoAWQ
248
+ cd AutoAWQ
249
+ pip3 install .
250
+ ```
251
+
252
+ ### Transformers example code (requires Transformers 4.35.0 and later)
253
+
254
+ ```python
255
+ from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer
256
+
257
+ model_name_or_path = "TheBloke/law-LLM-13B-AWQ"
258
+
259
+ tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
260
+ model = AutoModelForCausalLM.from_pretrained(
261
+ model_name_or_path,
262
+ low_cpu_mem_usage=True,
263
+ device_map="cuda:0"
264
+ )
265
+
266
+ # Using the text streamer to stream output one token at a time
267
+ streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
268
+
269
+ prompt = "Tell me about AI"
270
+ prompt_template=f'''[INST] <<SYS>>
271
+ {system_message}
272
+ <</SYS>>
273
+ {prompt} [/INST]
274
+ '''
275
+
276
+ # Convert prompt to tokens
277
+ tokens = tokenizer(
278
+ prompt_template,
279
+ return_tensors='pt'
280
+ ).input_ids.cuda()
281
+
282
+ generation_params = {
283
+ "do_sample": True,
284
+ "temperature": 0.7,
285
+ "top_p": 0.95,
286
+ "top_k": 40,
287
+ "max_new_tokens": 512,
288
+ "repetition_penalty": 1.1
289
+ }
290
+
291
+ # Generate streamed output, visible one token at a time
292
+ generation_output = model.generate(
293
+ tokens,
294
+ streamer=streamer,
295
+ **generation_params
296
+ )
297
+
298
+ # Generation without a streamer, which will include the prompt in the output
299
+ generation_output = model.generate(
300
+ tokens,
301
+ **generation_params
302
+ )
303
+
304
+ # Get the tokens from the output, decode them, print them
305
+ token_output = generation_output[0]
306
+ text_output = tokenizer.decode(token_output)
307
+ print("model.generate output: ", text_output)
308
+
309
+ # Inference is also possible via Transformers' pipeline
310
+ from transformers import pipeline
311
+
312
+ pipe = pipeline(
313
+ "text-generation",
314
+ model=model,
315
+ tokenizer=tokenizer,
316
+ **generation_params
317
+ )
318
+
319
+ pipe_output = pipe(prompt_template)[0]['generated_text']
320
+ print("pipeline output: ", pipe_output)
321
+
322
+ ```
323
+ <!-- README_AWQ.md-use-from-python end -->
324
+
325
+ <!-- README_AWQ.md-compatibility start -->
326
+ ## Compatibility
327
+
328
+ The files provided are tested to work with:
329
+
330
+ - [text-generation-webui](https://github.com/oobabooga/text-generation-webui) using `Loader: AutoAWQ`.
331
+ - [vLLM](https://github.com/vllm-project/vllm) version 0.2.0 and later.
332
+ - [Hugging Face Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference) version 1.1.0 and later.
333
+ - [Transformers](https://huggingface.co/docs/transformers) version 4.35.0 and later.
334
+ - [AutoAWQ](https://github.com/casper-hansen/AutoAWQ) version 0.1.1 and later.
335
+
336
+ <!-- README_AWQ.md-compatibility end -->
337
+
338
+ <!-- footer start -->
339
+ <!-- 200823 -->
340
+ ## Discord
341
+
342
+ For further support, and discussions on these models and AI in general, join us at:
343
+
344
+ [TheBloke AI's Discord server](https://discord.gg/theblokeai)
345
+
346
+ ## Thanks, and how to contribute
347
+
348
+ Thanks to the [chirper.ai](https://chirper.ai) team!
349
+
350
+ Thanks to Clay from [gpus.llm-utils.org](llm-utils)!
351
+
352
+ I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training.
353
+
354
+ If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects.
355
+
356
+ Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.
357
+
358
+ * Patreon: https://patreon.com/TheBlokeAI
359
+ * Ko-Fi: https://ko-fi.com/TheBlokeAI
360
+
361
+ **Special thanks to**: Aemon Algiz.
362
+
363
+ **Patreon special mentions**: Michael Levine, 阿明, Trailburnt, Nikolai Manek, John Detwiler, Randy H, Will Dee, Sebastain Graf, NimbleBox.ai, Eugene Pentland, Emad Mostaque, Ai Maven, Jim Angel, Jeff Scroggin, Michael Davis, Manuel Alberto Morcote, Stephen Murray, Robert, Justin Joy, Luke @flexchar, Brandon Frisco, Elijah Stavena, S_X, Dan Guido, Undi ., Komninos Chatzipapas, Shadi, theTransient, Lone Striker, Raven Klaugh, jjj, Cap'n Zoog, Michel-Marie MAUDET (LINAGORA), Matthew Berman, David, Fen Risland, Omer Bin Jawed, Luke Pendergrass, Kalila, OG, Erik Bjäreholt, Rooh Singh, Joseph William Delisle, Dan Lewis, TL, John Villwock, AzureBlack, Brad, Pedro Madruga, Caitlyn Gatomon, K, jinyuan sun, Mano Prime, Alex, Jeffrey Morgan, Alicia Loh, Illia Dulskyi, Chadd, transmissions 11, fincy, Rainer Wilmers, ReadyPlayerEmma, knownsqashed, Mandus, biorpg, Deo Leter, Brandon Phillips, SuperWojo, Sean Connelly, Iucharbius, Jack West, Harry Royden McLaughlin, Nicholas, terasurfer, Vitor Caleffi, Duane Dunston, Johann-Peter Hartmann, David Ziegler, Olakabola, Ken Nordquist, Trenton Dambrowitz, Tom X Nguyen, Vadim, Ajan Kanaga, Leonard Tan, Clay Pascal, Alexandros Triantafyllidis, JM33133, Xule, vamX, ya boyyy, subjectnull, Talal Aujan, Alps Aficionado, wassieverse, Ari Malik, James Bentley, Woland, Spencer Kim, Michael Dempsey, Fred von Graf, Elle, zynix, William Richards, Stanislav Ovsiannikov, Edmond Seymore, Jonathan Leane, Martin Kemka, usrbinkat, Enrico Ros
364
+
365
+
366
+ Thank you to all my generous patrons and donaters!
367
+
368
+ And thank you again to a16z for their generous grant.
369
+
370
+ <!-- footer end -->
371
+
372
+ # Original model card: AdaptLLM's Law LLM 13B
373
+
374
+
375
+ # Adapt (Large) Language Models to Domains
376
+ This repo contains the domain-specific base model developed from **LLaMA-1-13B**, using the method in our paper [Adapting Large Language Models via Reading Comprehension](https://huggingface.co/papers/2309.09530).
377
+
378
+ We explore **continued pre-training on domain-specific corpora** for large language models. While this approach enriches LLMs with domain knowledge, it significantly hurts their prompting ability for question answering. Inspired by human learning via reading comprehension, we propose a simple method to **transform large-scale pre-training corpora into reading comprehension texts**, consistently improving prompting performance across tasks in biomedicine, finance, and law domains. **Our 7B model competes with much larger domain-specific models like BloombergGPT-50B**.
379
+
380
+ ### 🤗 We are currently working hard on developing models across different domains, scales and architectures! Please stay tuned! 🤗
381
+
382
+ **************************** **Updates** ****************************
383
+ * 12/19: Released our [13B base models](https://huggingface.co/AdaptLLM/law-LLM-13B) developed from LLaMA-1-13B.
384
+ * 12/8: Released our [chat models](https://huggingface.co/AdaptLLM/law-chat) developed from LLaMA-2-Chat-7B.
385
+ * 9/18: Released our [paper](https://huggingface.co/papers/2309.09530), [code](https://github.com/microsoft/LMOps), [data](https://huggingface.co/datasets/AdaptLLM/law-tasks), and [base models](https://huggingface.co/AdaptLLM/law-LLM) developed from LLaMA-1-7B.
386
+
387
+ ## Domain-Specific LLaMA-1
388
+ ### LLaMA-1-7B
389
+ In our paper, we develop three domain-specific models from LLaMA-1-7B, which are also available in Huggingface: [Biomedicine-LLM](https://huggingface.co/AdaptLLM/medicine-LLM), [Finance-LLM](https://huggingface.co/AdaptLLM/finance-LLM) and [Law-LLM](https://huggingface.co/AdaptLLM/law-LLM), the performances of our AdaptLLM compared to other domain-specific LLMs are:
390
+
391
+ <p align='center'>
392
+ <img src="https://cdn-uploads.huggingface.co/production/uploads/650801ced5578ef7e20b33d4/6efPwitFgy-pLTzvccdcP.png" width="700">
393
+ </p>
394
+
395
+ ### LLaMA-1-13B
396
+ Moreover, we scale up our base model to LLaMA-1-13B to see if **our method is similarly effective for larger-scale models**, and the results are consistently positive too: [Biomedicine-LLM-13B](https://huggingface.co/AdaptLLM/medicine-LLM-13B), [Finance-LLM-13B](https://huggingface.co/AdaptLLM/finance-LLM-13B) and [Law-LLM-13B](https://huggingface.co/AdaptLLM/law-LLM-13B).
397
+
398
+ ## Domain-Specific LLaMA-2-Chat
399
+ Our method is also effective for aligned models! LLaMA-2-Chat requires a [specific data format](https://huggingface.co/blog/llama2#how-to-prompt-llama-2), and our **reading comprehension can perfectly fit the data format** by transforming the reading comprehension into a multi-turn conversation. We have also open-sourced chat models in different domains: [Biomedicine-Chat](https://huggingface.co/AdaptLLM/medicine-chat), [Finance-Chat](https://huggingface.co/AdaptLLM/finance-chat) and [Law-Chat](https://huggingface.co/AdaptLLM/law-chat)
400
+
401
+ For example, to chat with the law model:
402
+ ```python
403
+ from transformers import AutoModelForCausalLM, AutoTokenizer
404
+
405
+ model = AutoModelForCausalLM.from_pretrained("AdaptLLM/law-chat")
406
+ tokenizer = AutoTokenizer.from_pretrained("AdaptLLM/law-chat", use_fast=False)
407
+
408
+ # Put your input here:
409
+ user_input = '''Question: Which of the following is false about ex post facto laws?
410
+ Options:
411
+ - They make criminal an act that was innocent when committed.
412
+ - They prescribe greater punishment for an act than was prescribed when it was done.
413
+ - They increase the evidence required to convict a person than when the act was done.
414
+ - They alter criminal offenses or punishment in a substantially prejudicial manner for the purpose of punishing a person for some past activity.
415
+
416
+ Please provide your choice first and then provide explanations if possible.'''
417
+
418
+ # We use the prompt template of LLaMA-2-Chat demo
419
+ prompt = f"<s>[INST] <<SYS>>\nYou are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.\n\nIf a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.\n<</SYS>>\n\n{user_input} [/INST]"
420
+
421
+ inputs = tokenizer(prompt, return_tensors="pt", add_special_tokens=False).input_ids.to(model.device)
422
+ outputs = model.generate(input_ids=inputs, max_length=4096)[0]
423
+
424
+ answer_start = int(inputs.shape[-1])
425
+ pred = tokenizer.decode(outputs[answer_start:], skip_special_tokens=True)
426
+
427
+ print(f'### User Input:\n{user_input}\n\n### Assistant Output:\n{pred}')
428
+ ```
429
+
430
+ ## Domain-Specific Tasks
431
+ To easily reproduce our results, we have uploaded the filled-in zero/few-shot input instructions and output completions of each domain-specific task: [biomedicine-tasks](https://huggingface.co/datasets/AdaptLLM/medicine-tasks), [finance-tasks](https://huggingface.co/datasets/AdaptLLM/finance-tasks), and [law-tasks](https://huggingface.co/datasets/AdaptLLM/law-tasks).
432
+
433
+ **Note:** those filled-in instructions are specifically tailored for models before alignment and do NOT fit for the specific data format required for chat models.
434
+
435
+ ## Citation
436
+ If you find our work helpful, please cite us:
437
+ ```bibtex
438
+ @article{adaptllm,
439
+ title = {Adapting Large Language Models via Reading Comprehension},
440
+ author = {Daixuan Cheng and Shaohan Huang and Furu Wei},
441
+ journal = {CoRR},
442
+ volume = {abs/2309.09530},
443
+ year = {2023}
444
+ }
445
+ ```