TheBloke commited on
Commit
48cc64f
1 Parent(s): 653f05b

Update for Transformers GPTQ support

Browse files
README.md CHANGED
@@ -5,17 +5,20 @@ model_type: llama
5
  ---
6
 
7
  <!-- header start -->
8
- <div style="width: 100%;">
9
- <img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
 
10
  </div>
11
  <div style="display: flex; justify-content: space-between; width: 100%;">
12
  <div style="display: flex; flex-direction: column; align-items: flex-start;">
13
- <p><a href="https://discord.gg/theblokeai">Chat & support: my new Discord server</a></p>
14
  </div>
15
  <div style="display: flex; flex-direction: column; align-items: flex-end;">
16
- <p><a href="https://www.patreon.com/TheBlokeAI">Want to contribute? TheBloke's Patreon page</a></p>
17
  </div>
18
  </div>
 
 
19
  <!-- header end -->
20
 
21
  # Tim Dettmers' Guanaco 65B GPTQ
@@ -160,6 +163,7 @@ The files provided will work with AutoGPTQ (CUDA and Triton modes), GPTQ-for-LLa
160
  ExLlama works with Llama models in 4-bit. Please see the Provided Files table above for per-file compatibility.
161
 
162
  <!-- footer start -->
 
163
  ## Discord
164
 
165
  For further support, and discussions on these models and AI in general, join us at:
@@ -179,19 +183,22 @@ Donaters will get priority support on any and all AI/LLM/model questions and req
179
  * Patreon: https://patreon.com/TheBlokeAI
180
  * Ko-Fi: https://ko-fi.com/TheBlokeAI
181
 
182
- **Special thanks to**: Luke from CarbonQuill, Aemon Algiz.
 
 
183
 
184
- **Patreon special mentions**: Space Cruiser, Nikolai Manek, Sam, Chris McCloskey, Rishabh Srivastava, Kalila, Spiking Neurons AB, Khalefa Al-Ahmad, WelcomeToTheClub, Chadd, Lone Striker, Viktor Bowallius, Edmond Seymore, Ai Maven, Chris Smitley, Dave, Alexandros Triantafyllidis, Luke @flexchar, Elle, ya boyyy, Talal Aujan, Alex , Jonathan Leane, Deep Realms, Randy H, subjectnull, Preetika Verma, Joseph William Delisle, Michael Levine, chris gileta, K, Oscar Rangel, LangChain4j, Trenton Dambrowitz, Eugene Pentland, Johann-Peter Hartmann, Femi Adebogun, Illia Dulskyi, senxiiz, Daniel P. Andersen, Sean Connelly, Artur Olbinski, RoA, Mano Prime, Derek Yates, Raven Klaugh, David Flickinger, Willem Michiel, Pieter, Willian Hasse, vamX, Luke Pendergrass, webtim, Ghost , Rainer Wilmers, Nathan LeClaire, Will Dee, Cory Kujawski, John Detwiler, Fred von Graf, biorpg, Iucharbius , Imad Khwaja, Pierre Kircher, terasurfer , Asp the Wyvern, John Villwock, theTransient, zynix , Gabriel Tamborski, Fen Risland, Gabriel Puliatti, Matthew Berman, Pyrater, SuperWojo, Stephen Murray, Karl Bernard, Ajan Kanaga, Greatston Gnanesh, Junyu Yang.
185
 
186
  Thank you to all my generous patrons and donaters!
187
 
 
 
188
  <!-- footer end -->
189
 
190
  # Original model card: Tim Dettmers' Guanaco 65B
191
 
192
  # Guanaco Models Based on LLaMA
193
 
194
- | [Paper](https://arxiv.org/abs/2305.14314) | [Code](https://github.com/artidoro/qlora) | [Demo](https://huggingface.co/spaces/uwnlp/guanaco-playground-tgi) |
195
 
196
  **The Guanaco models are open-source finetuned chatbots obtained through 4-bit QLoRA tuning of LLaMA base models on the OASST1 dataset. They are available in 7B, 13B, 33B, and 65B parameter sizes.**
197
 
@@ -201,18 +208,18 @@ Thank you to all my generous patrons and donaters!
201
  - **Competitive with commercial chatbot systems on the Vicuna and OpenAssistant benchmarks** (ChatGPT and BARD) according to human and GPT-4 raters. We note that the relative performance on tasks not covered in these benchmarks could be very different. In addition, commercial systems evolve over time (we used outputs from the March 2023 version of the models).
202
  - **Available open-source for research purposes**. Guanaco models allow *cheap* and *local* experimentation with high-quality chatbot systems.
203
  - **Replicable and efficient training procedure** that can be extended to new use cases. Guanaco training scripts are available in the [QLoRA repo](https://github.com/artidoro/qlora).
204
- - **Rigorous comparison to 16-bit methods** (both 16-bit full-finetuning and LoRA) in [our paper](https://arxiv.org/abs/2305.14314) demonstrates the effectiveness of 4-bit QLoRA finetuning.
205
  - **Lightweight** checkpoints which only contain adapter weights.
206
 
207
  ## License and Intended Use
208
- Guanaco adapter weights are available under Apache 2 license. Note the use of the Guanaco adapter weights, requires access to the LLaMA model weighs.
209
- Guanaco is based on LLaMA and therefore should be used according to the LLaMA license.
210
 
211
  ## Usage
212
  Here is an example of how you would load Guanaco 7B in 4-bits:
213
  ```python
214
  import torch
215
- from peft import PeftModel
216
  from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
217
 
218
  model_name = "huggyllama/llama-7b"
@@ -254,7 +261,7 @@ A chat between a curious human and an artificial intelligence assistant. The ass
254
  ```
255
 
256
 
257
- ## Current Inference Limitations
258
  Currently, 4-bit inference is slow. We recommend loading in 16 bits if inference speed is a concern. We are actively working on releasing efficient 4-bit inference kernels.
259
 
260
  Below is how you would load the model in 16 bits:
@@ -280,18 +287,18 @@ tokenizer = AutoTokenizer.from_pretrained(model_name)
280
 
281
  **Finetuning Data**: Guanaco is finetuned on OASST1. The exact dataset is available at [timdettmers/openassistant-guanaco](https://huggingface.co/datasets/timdettmers/openassistant-guanaco).
282
 
283
- **Languages**: The OASST1 dataset is multilingual (see [the paper](https://arxiv.org/abs/2304.07327) for details) and as such Guanaco responds to user queries in different languages. We note, however, that OASST1 is heavy in high-resource languages. In addition, human evaluation of Guanaco was only performed in English and based on qualitative analysis we observed degradation in performance in other languages.
284
 
285
  Next, we describe Training and Evaluation details.
286
 
287
  ### Training
288
- Guanaco models are the result of 4-bit QLoRA supervised finetuning on the OASST1 dataset.
289
 
290
  All models use NormalFloat4 datatype for the base model and LoRA adapters on all linear layers with BFloat16 as computation datatype. We set LoRA $r=64$, $\alpha=16$. We also use Adam beta2 of 0.999, max grad norm of 0.3 and LoRA dropout of 0.1 for models up to 13B and 0.05 for 33B and 65B models.
291
- For the finetuning process, we use constant learning rate schedule and paged AdamW optimizer.
292
 
293
  ### Training hyperparameters
294
- Size| Dataset | Batch Size | Learning Rate | Max Steps | Sequence length
295
  ---|---|---|---|---|---
296
  7B | OASST1 | 16 | 2e-4 | 1875 | 512
297
  13B | OASST1 | 16 | 2e-4 | 1875 | 512
@@ -299,39 +306,39 @@ Size| Dataset | Batch Size | Learning Rate | Max Steps | Sequence length
299
  65B | OASST1 | 16 | 1e-4 | 1875 | 512
300
 
301
  ### Evaluation
302
- We test generative language capabilities through both automated and human evaluations. This second set of evaluations relies on queries curated by humans and aims at measuring the quality of model responses. We use the Vicuna and OpenAssistant datasets with 80 and 953 prompts respectively.
303
 
304
  In both human and automated evaluations, for each prompt, raters compare all pairs of responses across the models considered. For human raters we randomize the order of the systems, for GPT-4 we evaluate with both orders.
305
 
306
-
307
  Benchmark | Vicuna | | Vicuna | | OpenAssistant | | -
308
  -----------|----|-----|--------|---|---------------|---|---
309
  Prompts | 80 | | 80 | | 953 | |
310
- Judge | Human | | GPT-4 | | GPT-4 | |
311
- Model | Elo | Rank | Elo | Rank | Elo | Rank | **Median Rank**
312
- GPT-4 | 1176 | 1 | 1348 | 1 | 1294 | 1 | 1
313
- Guanaco-65B | 1023 | 2 | 1022 | 2 | 1008 | 3 | 2
314
- Guanaco-33B | 1009 | 4 | 992 | 3 | 1002 | 4 | 4
315
- ChatGPT-3.5 Turbo | 916 | 7 | 966 | 5 | 1015 | 2 | 5
316
- Vicuna-13B | 984 | 5 | 974 | 4 | 936 | 5 | 5
317
- Guanaco-13B | 975 | 6 | 913 | 6 | 885 | 6 | 6
318
- Guanaco-7B | 1010 | 3 | 879 | 8 | 860 | 7 | 7
319
- Bard | 909 | 8 | 902 | 7 | - | - | 8
320
 
321
 
322
  We also use the MMLU benchmark to measure performance on a range of language understanding tasks. This is a multiple-choice benchmark covering 57 tasks including elementary mathematics, US history, computer science, law, and more. We report 5-shot test accuracy.
323
 
324
- Dataset | 7B | 13B | 33B | 65B
325
  ---|---|---|---|---
326
- LLaMA no tuning | 35.1 | 46.9 | 57.8 | 63.4
327
- Self-Instruct | 36.4 | 33.3 | 53.0 | 56.7
328
- Longform | 32.1 | 43.2 | 56.6 | 59.7
329
- Chip2 | 34.5 | 41.6 | 53.6 | 59.8
330
- HH-RLHF | 34.9 | 44.6 | 55.8 | 60.1
331
- Unnatural Instruct | 41.9 | 48.1 | 57.3 | 61.3
332
- OASST1 (Guanaco) | 36.6 | 46.4 | 57.0 | 62.2
333
- Alpaca | 38.8 | 47.8 | 57.3 | 62.5
334
- FLAN v2 | 44.5 | 51.4 | 59.2 | 63.9
335
 
336
  ## Risks and Biases
337
  The model can produce factually incorrect output, and should not be relied on to produce factually accurate information. The model was trained on various public datasets; it is possible that this model could generate lewd, biased, or otherwise offensive outputs.
 
5
  ---
6
 
7
  <!-- header start -->
8
+ <!-- 200823 -->
9
+ <div style="width: auto; margin-left: auto; margin-right: auto">
10
+ <img src="https://i.imgur.com/EBdldam.jpg" alt="TheBlokeAI" style="width: 100%; min-width: 400px; display: block; margin: auto;">
11
  </div>
12
  <div style="display: flex; justify-content: space-between; width: 100%;">
13
  <div style="display: flex; flex-direction: column; align-items: flex-start;">
14
+ <p style="margin-top: 0.5em; margin-bottom: 0em;"><a href="https://discord.gg/theblokeai">Chat & support: TheBloke's Discord server</a></p>
15
  </div>
16
  <div style="display: flex; flex-direction: column; align-items: flex-end;">
17
+ <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>
18
  </div>
19
  </div>
20
+ <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>
21
+ <hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
22
  <!-- header end -->
23
 
24
  # Tim Dettmers' Guanaco 65B GPTQ
 
163
  ExLlama works with Llama models in 4-bit. Please see the Provided Files table above for per-file compatibility.
164
 
165
  <!-- footer start -->
166
+ <!-- 200823 -->
167
  ## Discord
168
 
169
  For further support, and discussions on these models and AI in general, join us at:
 
183
  * Patreon: https://patreon.com/TheBlokeAI
184
  * Ko-Fi: https://ko-fi.com/TheBlokeAI
185
 
186
+ **Special thanks to**: Aemon Algiz.
187
+
188
+ **Patreon special mentions**: Sam, theTransient, Jonathan Leane, Steven Wood, webtim, Johann-Peter Hartmann, Geoffrey Montalvo, Gabriel Tamborski, Willem Michiel, John Villwock, Derek Yates, Mesiah Bishop, Eugene Pentland, Pieter, Chadd, Stephen Murray, Daniel P. Andersen, terasurfer, Brandon Frisco, Thomas Belote, Sid, Nathan LeClaire, Magnesian, Alps Aficionado, Stanislav Ovsiannikov, Alex, Joseph William Delisle, Nikolai Manek, Michael Davis, Junyu Yang, K, J, Spencer Kim, Stefan Sabev, Olusegun Samson, transmissions 11, Michael Levine, Cory Kujawski, Rainer Wilmers, zynix, Kalila, Luke @flexchar, Ajan Kanaga, Mandus, vamX, Ai Maven, Mano Prime, Matthew Berman, subjectnull, Vitor Caleffi, Clay Pascal, biorpg, alfie_i, 阿明, Jeffrey Morgan, ya boyyy, Raymond Fosdick, knownsqashed, Olakabola, Leonard Tan, ReadyPlayerEmma, Enrico Ros, Dave, Talal Aujan, Illia Dulskyi, Sean Connelly, senxiiz, Artur Olbinski, Elle, Raven Klaugh, Fen Risland, Deep Realms, Imad Khwaja, Fred von Graf, Will Dee, usrbinkat, SuperWojo, Alexandros Triantafyllidis, Swaroop Kallakuri, Dan Guido, John Detwiler, Pedro Madruga, Iucharbius, Viktor Bowallius, Asp the Wyvern, Edmond Seymore, Trenton Dambrowitz, Space Cruiser, Spiking Neurons AB, Pyrater, LangChain4j, Tony Hughes, Kacper Wikieł, Rishabh Srivastava, David Ziegler, Luke Pendergrass, Andrey, Gabriel Puliatti, Lone Striker, Sebastain Graf, Pierre Kircher, Randy H, NimbleBox.ai, Vadim, danny, Deo Leter
189
 
 
190
 
191
  Thank you to all my generous patrons and donaters!
192
 
193
+ And thank you again to a16z for their generous grant.
194
+
195
  <!-- footer end -->
196
 
197
  # Original model card: Tim Dettmers' Guanaco 65B
198
 
199
  # Guanaco Models Based on LLaMA
200
 
201
+ | [Paper](https://arxiv.org/abs/2305.14314) | [Code](https://github.com/artidoro/qlora) | [Demo](https://huggingface.co/spaces/uwnlp/guanaco-playground-tgi) |
202
 
203
  **The Guanaco models are open-source finetuned chatbots obtained through 4-bit QLoRA tuning of LLaMA base models on the OASST1 dataset. They are available in 7B, 13B, 33B, and 65B parameter sizes.**
204
 
 
208
  - **Competitive with commercial chatbot systems on the Vicuna and OpenAssistant benchmarks** (ChatGPT and BARD) according to human and GPT-4 raters. We note that the relative performance on tasks not covered in these benchmarks could be very different. In addition, commercial systems evolve over time (we used outputs from the March 2023 version of the models).
209
  - **Available open-source for research purposes**. Guanaco models allow *cheap* and *local* experimentation with high-quality chatbot systems.
210
  - **Replicable and efficient training procedure** that can be extended to new use cases. Guanaco training scripts are available in the [QLoRA repo](https://github.com/artidoro/qlora).
211
+ - **Rigorous comparison to 16-bit methods** (both 16-bit full-finetuning and LoRA) in [our paper](https://arxiv.org/abs/2305.14314) demonstrates the effectiveness of 4-bit QLoRA finetuning.
212
  - **Lightweight** checkpoints which only contain adapter weights.
213
 
214
  ## License and Intended Use
215
+ Guanaco adapter weights are available under Apache 2 license. Note the use of the Guanaco adapter weights, requires access to the LLaMA model weighs.
216
+ Guanaco is based on LLaMA and therefore should be used according to the LLaMA license.
217
 
218
  ## Usage
219
  Here is an example of how you would load Guanaco 7B in 4-bits:
220
  ```python
221
  import torch
222
+ from peft import PeftModel
223
  from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
224
 
225
  model_name = "huggyllama/llama-7b"
 
261
  ```
262
 
263
 
264
+ ## Current Inference Limitations
265
  Currently, 4-bit inference is slow. We recommend loading in 16 bits if inference speed is a concern. We are actively working on releasing efficient 4-bit inference kernels.
266
 
267
  Below is how you would load the model in 16 bits:
 
287
 
288
  **Finetuning Data**: Guanaco is finetuned on OASST1. The exact dataset is available at [timdettmers/openassistant-guanaco](https://huggingface.co/datasets/timdettmers/openassistant-guanaco).
289
 
290
+ **Languages**: The OASST1 dataset is multilingual (see [the paper](https://arxiv.org/abs/2304.07327) for details) and as such Guanaco responds to user queries in different languages. We note, however, that OASST1 is heavy in high-resource languages. In addition, human evaluation of Guanaco was only performed in English and based on qualitative analysis we observed degradation in performance in other languages.
291
 
292
  Next, we describe Training and Evaluation details.
293
 
294
  ### Training
295
+ Guanaco models are the result of 4-bit QLoRA supervised finetuning on the OASST1 dataset.
296
 
297
  All models use NormalFloat4 datatype for the base model and LoRA adapters on all linear layers with BFloat16 as computation datatype. We set LoRA $r=64$, $\alpha=16$. We also use Adam beta2 of 0.999, max grad norm of 0.3 and LoRA dropout of 0.1 for models up to 13B and 0.05 for 33B and 65B models.
298
+ For the finetuning process, we use constant learning rate schedule and paged AdamW optimizer.
299
 
300
  ### Training hyperparameters
301
+ Size| Dataset | Batch Size | Learning Rate | Max Steps | Sequence length
302
  ---|---|---|---|---|---
303
  7B | OASST1 | 16 | 2e-4 | 1875 | 512
304
  13B | OASST1 | 16 | 2e-4 | 1875 | 512
 
306
  65B | OASST1 | 16 | 1e-4 | 1875 | 512
307
 
308
  ### Evaluation
309
+ We test generative language capabilities through both automated and human evaluations. This second set of evaluations relies on queries curated by humans and aims at measuring the quality of model responses. We use the Vicuna and OpenAssistant datasets with 80 and 953 prompts respectively.
310
 
311
  In both human and automated evaluations, for each prompt, raters compare all pairs of responses across the models considered. For human raters we randomize the order of the systems, for GPT-4 we evaluate with both orders.
312
 
313
+
314
  Benchmark | Vicuna | | Vicuna | | OpenAssistant | | -
315
  -----------|----|-----|--------|---|---------------|---|---
316
  Prompts | 80 | | 80 | | 953 | |
317
+ Judge | Human | | GPT-4 | | GPT-4 | |
318
+ Model | Elo | Rank | Elo | Rank | Elo | Rank | **Median Rank**
319
+ GPT-4 | 1176 | 1 | 1348 | 1 | 1294 | 1 | 1
320
+ Guanaco-65B | 1023 | 2 | 1022 | 2 | 1008 | 3 | 2
321
+ Guanaco-33B | 1009 | 4 | 992 | 3 | 1002 | 4 | 4
322
+ ChatGPT-3.5 Turbo | 916 | 7 | 966 | 5 | 1015 | 2 | 5
323
+ Vicuna-13B | 984 | 5 | 974 | 4 | 936 | 5 | 5
324
+ Guanaco-13B | 975 | 6 | 913 | 6 | 885 | 6 | 6
325
+ Guanaco-7B | 1010 | 3 | 879 | 8 | 860 | 7 | 7
326
+ Bard | 909 | 8 | 902 | 7 | - | - | 8
327
 
328
 
329
  We also use the MMLU benchmark to measure performance on a range of language understanding tasks. This is a multiple-choice benchmark covering 57 tasks including elementary mathematics, US history, computer science, law, and more. We report 5-shot test accuracy.
330
 
331
+ Dataset | 7B | 13B | 33B | 65B
332
  ---|---|---|---|---
333
+ LLaMA no tuning | 35.1 | 46.9 | 57.8 | 63.4
334
+ Self-Instruct | 36.4 | 33.3 | 53.0 | 56.7
335
+ Longform | 32.1 | 43.2 | 56.6 | 59.7
336
+ Chip2 | 34.5 | 41.6 | 53.6 | 59.8
337
+ HH-RLHF | 34.9 | 44.6 | 55.8 | 60.1
338
+ Unnatural Instruct | 41.9 | 48.1 | 57.3 | 61.3
339
+ OASST1 (Guanaco) | 36.6 | 46.4 | 57.0 | 62.2
340
+ Alpaca | 38.8 | 47.8 | 57.3 | 62.5
341
+ FLAN v2 | 44.5 | 51.4 | 59.2 | 63.9
342
 
343
  ## Risks and Biases
344
  The model can produce factually incorrect output, and should not be relied on to produce factually accurate information. The model was trained on various public datasets; it is possible that this model could generate lewd, biased, or otherwise offensive outputs.
config.json CHANGED
@@ -1,24 +1,34 @@
1
  {
2
- "_name_or_path": "/workspace/models/huggyllama_llama-65b",
3
- "architectures": [
4
- "LlamaForCausalLM"
5
- ],
6
- "bos_token_id": 1,
7
- "eos_token_id": 2,
8
- "hidden_act": "silu",
9
- "hidden_size": 8192,
10
- "initializer_range": 0.02,
11
- "intermediate_size": 22016,
12
- "max_position_embeddings": 2048,
13
- "max_sequence_length": 2048,
14
- "model_type": "llama",
15
- "num_attention_heads": 64,
16
- "num_hidden_layers": 80,
17
- "pad_token_id": 0,
18
- "rms_norm_eps": 1e-05,
19
- "tie_word_embeddings": false,
20
- "torch_dtype": "float16",
21
- "transformers_version": "4.29.2",
22
- "use_cache": true,
23
- "vocab_size": 32000
 
 
 
 
 
 
 
 
 
 
24
  }
 
1
  {
2
+ "_name_or_path": "/workspace/models/huggyllama_llama-65b",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "bos_token_id": 1,
7
+ "eos_token_id": 2,
8
+ "hidden_act": "silu",
9
+ "hidden_size": 8192,
10
+ "initializer_range": 0.02,
11
+ "intermediate_size": 22016,
12
+ "max_position_embeddings": 2048,
13
+ "max_sequence_length": 2048,
14
+ "model_type": "llama",
15
+ "num_attention_heads": 64,
16
+ "num_hidden_layers": 80,
17
+ "pad_token_id": 0,
18
+ "rms_norm_eps": 1e-05,
19
+ "tie_word_embeddings": false,
20
+ "torch_dtype": "float16",
21
+ "transformers_version": "4.29.2",
22
+ "use_cache": true,
23
+ "vocab_size": 32000,
24
+ "quantization_config": {
25
+ "bits": 4,
26
+ "group_size": -1,
27
+ "damp_percent": 0.01,
28
+ "desc_act": true,
29
+ "sym": true,
30
+ "true_sequential": true,
31
+ "model_file_base_name": "model",
32
+ "quant_method": "gptq"
33
+ }
34
  }
Guanaco-65B-GPTQ-4bit.act-order.safetensors → model.safetensors RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:df703858c425f68224d46d40583c21f3db470d789de282974b37c8d00e193874
3
- size 33489332352
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c1721c347f2a86fef0aea172860e5643d0c43adf582ebf7c6955bd94ac9db3f3
3
+ size 33489332408
quantize_config.json CHANGED
@@ -1,8 +1,9 @@
1
  {
2
- "bits": 4,
3
- "group_size": -1,
4
- "damp_percent": 0.01,
5
- "desc_act": true,
6
- "sym": true,
7
- "true_sequential": true
 
8
  }
 
1
  {
2
+ "bits": 4,
3
+ "group_size": -1,
4
+ "damp_percent": 0.01,
5
+ "desc_act": true,
6
+ "sym": true,
7
+ "true_sequential": true,
8
+ "model_file_base_name": "model"
9
  }