jaspercatapang commited on
Commit
42ad143
1 Parent(s): 7b78087

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +92 -2
README.md CHANGED
@@ -11,6 +11,96 @@ tags:
11
  <img src="https://drive.google.com/uc?export=view&id=1D8wxXkS1nsq3uqbOzOLwgx1cLJhY1nvN" alt="GodziLLa2-70B">
12
  Released August 11, 2023
13
 
14
- <b>Model card coming soon!</b>
 
 
15
 
16
- GodziLLa-30B can be found [here](https://huggingface.co/MayaPH/GodziLLa-30B).
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
11
  <img src="https://drive.google.com/uc?export=view&id=1D8wxXkS1nsq3uqbOzOLwgx1cLJhY1nvN" alt="GodziLLa2-70B">
12
  Released August 11, 2023
13
 
14
+ ## Model Description
15
+ GodziLLa 2 70B is an experimental combination of various proprietary LoRAs from Maya Philippines and [Guanaco LLaMA 2 1K dataset](https://huggingface.co/datasets/mlabonne/guanaco-llama2-1k), with LLaMA 2 70B. This model's primary purpose is to stress test the limitations of composite, instruction-following LLMs and observe its performance with respect to other LLMs available on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard). This model debuted in the leaderboard at rank #5 (August 17, 2023).
16
+ ![Godzilla Happy GIF](https://i.pinimg.com/originals/81/3a/e0/813ae09a30f0bc44130cd2c834fe2eba.gif)
17
 
18
+ ## Open LLM Leaderboard Metrics
19
+ | Metric | Value |
20
+ |-----------------------|-------|
21
+ | MMLU (5-shot) | 69.88 |
22
+ | ARC (25-shot) | 71.42 |
23
+ | HellaSwag (10-shot) | 87.53 |
24
+ | TruthfulQA (0-shot) | 61.54 |
25
+ | Average | 72.59 |
26
+
27
+ According to the leaderboard description, here are the benchmarks used for the evaluation:
28
+ - [MMLU](https://arxiv.org/abs/2009.03300) (5-shot) - a test to measure a text model’s multitask accuracy. The test covers 57 tasks including elementary mathematics, US history, computer science, law, and more.
29
+ - [AI2 Reasoning Challenge](https://arxiv.org/abs/1803.05457) -ARC- (25-shot) - a set of grade-school science questions.
30
+ - [HellaSwag](https://arxiv.org/abs/1905.07830) (10-shot) - a test of commonsense inference, which is easy for humans (~95%) but challenging for SOTA models.
31
+ - [TruthfulQA](https://arxiv.org/abs/2109.07958) (0-shot) - a test to measure a model’s propensity to reproduce falsehoods commonly found online.
32
+
33
+ ## Leaderboard Highlights (as of August 17, 2023)
34
+ - Godzilla 2 70B ranks #5, worldwide, in the Open LLM Leaderboard.
35
+ - Godzilla 2 70B ranks #3 in the ARC challenge.
36
+ - Godzilla 2 70B ranks #6 in the TruthfulQA benchmark.
37
+ - *Godzilla 2 70B beats GPT-3.5 (ChatGPT) in terms of average performance and the HellaSwag benchmark (87.53 > 85.5).
38
+ - *Godzilla 2 70B outperforms GPT-3.5 (ChatGPT) and GPT-4 on the TruthfulQA benchmark (61.54 for G2-70B, 47 for GPT-3.5, 59 for GPT-4).
39
+ - *Godzilla 2 70B is on par with GPT-3.5 (ChatGPT) on the MMLU benchmark (<0.12%).
40
+ *Based on a [leaderboard clone](https://huggingface.co/spaces/gsaivinay/open_llm_leaderboard) with GPT-3.5 and GPT-4 included.
41
+
42
+ ### Reproducing Evaluation Results
43
+
44
+ Install LM Evaluation Harness:
45
+ ```
46
+ # clone repository
47
+ git clone https://github.com/EleutherAI/lm-evaluation-harness.git
48
+ # change to repo directory
49
+ cd lm-evaluation-harness
50
+ # check out the correct commit
51
+ git checkout b281b0921b636bc36ad05c0b0b0763bd6dd43463
52
+ # install
53
+ pip install -e .
54
+ ```
55
+ Each task was evaluated on a single A100 80GB GPU.
56
+
57
+ ARC:
58
+ ```
59
+ python main.py --model hf-causal-experimental --model_args pretrained=garage-bAInd/Platypus2-70B-instruct --tasks arc_challenge --batch_size 1 --no_cache --write_out --output_path results/Platypus2-70B-instruct/arc_challenge_25shot.json --device cuda --num_fewshot 25
60
+ ```
61
+
62
+ HellaSwag:
63
+ ```
64
+ python main.py --model hf-causal-experimental --model_args pretrained=garage-bAInd/Platypus2-70B-instruct --tasks hellaswag --batch_size 1 --no_cache --write_out --output_path results/Platypus2-70B-instruct/hellaswag_10shot.json --device cuda --num_fewshot 10
65
+ ```
66
+
67
+ MMLU:
68
+ ```
69
+ python main.py --model hf-causal-experimental --model_args pretrained=garage-bAInd/Platypus2-70B-instruct --tasks hendrycksTest-* --batch_size 1 --no_cache --write_out --output_path results/Platypus2-70B-instruct/mmlu_5shot.json --device cuda --num_fewshot 5
70
+ ```
71
+
72
+ TruthfulQA:
73
+ ```
74
+ python main.py --model hf-causal-experimental --model_args pretrained=garage-bAInd/Platypus2-70B-instruct --tasks truthfulqa_mc --batch_size 1 --no_cache --write_out --output_path results/Platypus2-70B-instruct/truthfulqa_0shot.json --device cuda
75
+ ```
76
+
77
+ ### Prompt Template
78
+ ```
79
+ ### Instruction:
80
+
81
+ <prompt> (without the <>)
82
+
83
+ ### Response:
84
+ ```
85
+
86
+ ## Ethical Considerations
87
+ When using GodziLLa 2 70B, it is important to consider the following ethical considerations:
88
+
89
+ 1. **Privacy and Security:** Avoid sharing sensitive personal information while interacting with the model. The model does not have privacy safeguards, so exercise caution when discussing personal or confidential matters.
90
+
91
+ 2. **Fairness and Bias:** The model's responses may reflect biases present in the training data. Be aware of potential biases and make an effort to evaluate responses critically and fairly.
92
+
93
+ 3. **Transparency:** The model operates as a predictive text generator based on patterns learned from the training data. The model's inner workings and the specific training data used are proprietary and not publicly available.
94
+
95
+ 4. **User Responsibility:** Users should take responsibility for their own decisions and not solely rely on the information provided by the model. Consult with the appropriate professionals or reliable sources for specific advice or recommendations.
96
+
97
+ 5. **NSFW Content:** The model is a merge of various datasets and LoRA adapters. It is highly likely that the resulting model contains uncensored content that may include, but is not limited to, violence, gore, explicit language, and sexual content. If you plan to further refine this model for safe/aligned usage, you are highly encouraged to implement guardrails along with it.
98
+
99
+ ## Further Information
100
+ For additional information or inquiries about GodziLLa 2 70B, please contact the Maya Philippines iOps Team via jasper.catapang@maya.ph.
101
+
102
+ ## Disclaimer
103
+ GodziLLa 2 70B is an AI language model from Maya Philippines. It is provided "as is" without warranty of any kind, express or implied. The model developers and Maya Philippines shall not be liable for any direct or indirect damages arising from the use of this model.
104
+
105
+ ## Acknowledgments
106
+ The development of GodziLLa 2 70B was made possible by Maya Philippines and the curation of the various proprietary datasets and creation of the different proprietary LoRA adapters. Special thanks to mlabonne for the Guanaco dataset found [here](https://huggingface.co/datasets/mlabonne/guanaco-llama2-1k).