willyninja30 commited on
Commit
c1d170d
1 Parent(s): 818dc71

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +11 -38
README.md CHANGED
@@ -24,55 +24,43 @@ The goal is to increase model quality on French and general topics.
24
  # **Aria 70B is based on Llama 2-70B-Chat-HF**
25
  Llama 2 is a collection of pretrained and fine-tuned generative text models ranging in scale from 7 billion to 70 billion parameters. This is the repository for the 70B fine-tuned model, optimized for dialogue use cases and converted for the Hugging Face Transformers format. Links to other models can be found in the index at the bottom.
26
 
27
- # *FINETUNING PROCESS - UPDATES**
28
 
29
- We trained the model on a high quality dataset with more than 50.000 rows of french language.
30
  ....
31
  ....
32
  # **Timing of training**
33
  2 Days using NVIDIA A10G and Amazon Web services Cloud Instance. We are grateful to Nvidia Inception program.
34
 
35
- We are also applying rope scalling as experimental approach used by several other Open source teams to increase context lenght of LLAMA 2 from 4,096 to over 6,000 tokens. This will allow the model to handle large files for data extraction.
36
  ## Model Details /
37
- *Note: Use of this model is governed by the Meta license. In order to download the model weights and tokenizer, please visit the [website](https://ai.meta.com/resources/models-and-libraries/llama-downloads/) and accept our License before requesting access here.*
38
 
39
 
40
- **Model Developers** Meta
41
 
42
- **Variations** Llama 2 comes in a range of parameter sizes — 7B, 13B, and 70B as well as pretrained and fine-tuned variations.
43
 
44
  **Input** Models input text only.
45
 
46
  **Output** Models generate text only.
47
 
48
- **Model Architecture** Llama 2 is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align to human preferences for helpfulness and safety.
49
 
50
  **License** A custom commercial license is available at: [https://ai.meta.com/resources/models-and-libraries/llama-downloads/](https://ai.meta.com/resources/models-and-libraries/llama-downloads/)
51
 
52
- **Research Paper** ["Llama-2: Open Foundation and Fine-tuned Chat Models"](arxiv.org/abs/2307.09288)
53
 
54
 
55
 
56
  **CO<sub>2</sub> emissions during pretraining.** Time: total GPU time required for training each model. Power Consumption: peak power capacity per GPU device for the GPUs used adjusted for power usage efficiency. 100% of the emissions are directly offset by Meta's sustainability program, and because we are openly releasing these models, the pretraining costs do not need to be incurred by others.
57
 
58
  ## Training Data
59
- **Overview** Llama 2 was pretrained on 2 trillion tokens of data from publicly available sources. The fine-tuning data includes publicly available instruction datasets, as well as over one million new human-annotated examples. Neither the pretraining nor the fine-tuning datasets include Meta user data.
60
 
61
- **Data Freshness** The pretraining data has a cutoff of September 2022, but some tuning data is more recent, up to July 2023.
62
 
63
- ## Evaluation Results
64
 
65
- In this section, we report the results for the Llama 1 and Llama 2 models on standard academic benchmarks.For all the evaluations, we use our internal evaluations library.
66
-
67
- |Model|Size|Code|Commonsense Reasoning|World Knowledge|Reading Comprehension|Math|MMLU|BBH|AGI Eval|
68
- |---|---|---|---|---|---|---|---|---|---|
69
- |Llama 1|7B|14.1|60.8|46.2|58.5|6.95|35.1|30.3|23.9|
70
- |Llama 1|13B|18.9|66.1|52.6|62.3|10.9|46.9|37.0|33.9|
71
- |Llama 1|33B|26.0|70.0|58.4|67.6|21.4|57.8|39.8|41.7|
72
- |Llama 1|65B|30.7|70.7|60.5|68.6|30.8|63.4|43.5|47.6|
73
- |Llama 2|7B|16.8|63.9|48.9|61.3|14.6|45.3|32.6|29.3|
74
- |Llama 2|13B|24.5|66.9|55.4|65.8|28.7|54.8|39.4|39.1|
75
- |Llama 2|70B|**37.5**|**71.9**|**63.6**|**69.4**|**35.2**|**68.9**|**51.2**|**54.2**|
76
 
77
  **Overall performance on grouped academic benchmarks.** *Code:* We report the average pass@1 scores of our models on HumanEval and MBPP. *Commonsense Reasoning:* We report the average of PIQA, SIQA, HellaSwag, WinoGrande, ARC easy and challenge, OpenBookQA, and CommonsenseQA. We report 7-shot results for CommonSenseQA and 0-shot results for all other benchmarks. *World Knowledge:* We evaluate the 5-shot performance on NaturalQuestions and TriviaQA and report the average. *Reading Comprehension:* For reading comprehension, we report the 0-shot average on SQuAD, QuAC, and BoolQ. *MATH:* We report the average of the GSM8K (8 shot) and MATH (4 shot) benchmarks at top 1.
78
 
@@ -98,19 +86,4 @@ In this section, we report the results for the Llama 1 and Llama 2 models on sta
98
  **Evaluation of fine-tuned LLMs on different safety datasets.** Same metric definitions as above.
99
 
100
  ## Ethical Considerations and Limitations
101
- Llama 2 is a new technology that carries risks with use. Testing conducted to date has been in English, and has not covered, nor could it cover all scenarios. For these reasons, as with all LLMs, Llama 2’s potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate, biased or other objectionable responses to user prompts. Therefore, before deploying any applications of Llama 2, developers should perform safety testing and tuning tailored to their specific applications of the model.
102
-
103
- Please see the Responsible Use Guide available at [https://ai.meta.com/llama/responsible-use-guide/](https://ai.meta.com/llama/responsible-use-guide)
104
-
105
- ## Reporting Issues
106
- Please report any software “bug,” or other problems with the models through one of the following means:
107
- - Reporting issues with the model: [github.com/facebookresearch/llama](http://github.com/facebookresearch/llama)
108
- - Reporting problematic content generated by the model: [developers.facebook.com/llama_output_feedback](http://developers.facebook.com/llama_output_feedback)
109
- - Reporting bugs and security concerns: [facebook.com/whitehat/info](http://facebook.com/whitehat/info)
110
-
111
- ## Llama Model Index
112
- |Model|Llama2|Llama2-hf|Llama2-chat|Llama2-chat-hf|
113
- |---|---|---|---|---|
114
- |7B| [Link](https://huggingface.co/llamaste/Llama-2-7b) | [Link](https://huggingface.co/llamaste/Llama-2-7b-hf) | [Link](https://huggingface.co/llamaste/Llama-2-7b-chat) | [Link](https://huggingface.co/llamaste/Llama-2-7b-chat-hf)|
115
- |13B| [Link](https://huggingface.co/llamaste/Llama-2-13b) | [Link](https://huggingface.co/llamaste/Llama-2-13b-hf) | [Link](https://huggingface.co/llamaste/Llama-2-13b-chat) | [Link](https://huggingface.co/llamaste/Llama-2-13b-hf)|
116
- |70B| [Link](https://huggingface.co/llamaste/Llama-2-70b) | [Link](https://huggingface.co/llamaste/Llama-2-70b-hf) | [Link](https://huggingface.co/llamaste/Llama-2-70b-chat) | [Link](https://huggingface.co/llamaste/Llama-2-70b-hf)|
 
24
  # **Aria 70B is based on Llama 2-70B-Chat-HF**
25
  Llama 2 is a collection of pretrained and fine-tuned generative text models ranging in scale from 7 billion to 70 billion parameters. This is the repository for the 70B fine-tuned model, optimized for dialogue use cases and converted for the Hugging Face Transformers format. Links to other models can be found in the index at the bottom.
26
 
27
+ # *FINETUNING PROCESS **
28
 
29
+ We trained the model on a high quality dataset with more than 50.000 rows of french language. The training took 2 days on Amazon Cloud Sagemaker powered by Nvidia GPUs.
30
  ....
31
  ....
32
  # **Timing of training**
33
  2 Days using NVIDIA A10G and Amazon Web services Cloud Instance. We are grateful to Nvidia Inception program.
34
 
35
+ We are also applying rope scalling as experimental approach used by several other Open source teams to increase context lenght of ARIA from 4,096 to over 6,000 tokens. This will allow the model to handle large files for data extraction. This is not active by default and you should add a line of code at parameters to activate rope scaling.
36
  ## Model Details /
37
+ *Note: Use of this model is governed by the Meta license because it's based on LLAMA 2. In order to download the model weights and tokenizer, please visit the [website](https://ai.meta.com/resources/models-and-libraries/llama-downloads/) and accept our License before requesting access here.*
38
 
39
 
40
+ **Model Developers** FARADAY
41
 
42
+ **Variations** ARIA comes in a range of parameter sizes — 7B, 40B (based on Falcon), and 70B finetuned on French language datasets.
43
 
44
  **Input** Models input text only.
45
 
46
  **Output** Models generate text only.
47
 
48
+ **Model Architecture** ARIA is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align to human preferences for helpfulness and safety.
49
 
50
  **License** A custom commercial license is available at: [https://ai.meta.com/resources/models-and-libraries/llama-downloads/](https://ai.meta.com/resources/models-and-libraries/llama-downloads/)
51
 
52
+ **Research Paper for LLAMA 2** ["Llama-2: Open Foundation and Fine-tuned Chat Models"](arxiv.org/abs/2307.09288)
53
 
54
 
55
 
56
  **CO<sub>2</sub> emissions during pretraining.** Time: total GPU time required for training each model. Power Consumption: peak power capacity per GPU device for the GPUs used adjusted for power usage efficiency. 100% of the emissions are directly offset by Meta's sustainability program, and because we are openly releasing these models, the pretraining costs do not need to be incurred by others.
57
 
58
  ## Training Data
 
59
 
60
+ **Overview** ARIA was trained over on 50.000 tokens of data from publicly available sources in French.
61
 
62
+ **Data Freshness** The pretraining data has a cutoff of September 2022, but some tuning data is more recent, up to August 2023.
63
 
 
 
 
 
 
 
 
 
 
 
 
64
 
65
  **Overall performance on grouped academic benchmarks.** *Code:* We report the average pass@1 scores of our models on HumanEval and MBPP. *Commonsense Reasoning:* We report the average of PIQA, SIQA, HellaSwag, WinoGrande, ARC easy and challenge, OpenBookQA, and CommonsenseQA. We report 7-shot results for CommonSenseQA and 0-shot results for all other benchmarks. *World Knowledge:* We evaluate the 5-shot performance on NaturalQuestions and TriviaQA and report the average. *Reading Comprehension:* For reading comprehension, we report the 0-shot average on SQuAD, QuAC, and BoolQ. *MATH:* We report the average of the GSM8K (8 shot) and MATH (4 shot) benchmarks at top 1.
66
 
 
86
  **Evaluation of fine-tuned LLMs on different safety datasets.** Same metric definitions as above.
87
 
88
  ## Ethical Considerations and Limitations
89
+ ARIA is a new technology that carries risks with use.