PhaedrusFlow commited on
Commit
599b45b
·
verified ·
1 Parent(s): 5125b9b

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +14 -26
README.md CHANGED
@@ -192,13 +192,13 @@ Meta developed and released the Meta Llama 3 family of large language models (LL
192
 
193
  **Model developers** Meta
194
 
195
- **Variations** Llama 3 comes in two sizes 8B and 70B parameters — in pre-trained and instruction tuned variants.
196
 
197
  **Input** Models input text only.
198
 
199
  **Output** Models generate text and code only.
200
 
201
- **Model Architecture** Llama 3 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 with human preferences for helpfulness and safety.
202
 
203
 
204
  <table>
@@ -219,7 +219,7 @@ Meta developed and released the Meta Llama 3 family of large language models (LL
219
  </td>
220
  </tr>
221
  <tr>
222
- <td rowspan="2" >Llama 3
223
  </td>
224
  <td rowspan="2" >A new mix of publicly available online data.
225
  </td>
@@ -247,11 +247,11 @@ Meta developed and released the Meta Llama 3 family of large language models (LL
247
  </table>
248
 
249
 
250
- **Llama 3 family of models**. Token counts refer to pretraining data only. Both the 8 and 70B versions use Grouped-Query Attention (GQA) for improved inference scalability.
251
 
252
  **Model Release Date** April 18, 2024.
253
 
254
- **Status** This is a static model trained on an offline dataset. Future versions of the tuned models will be released as we improve model safety with community feedback.
255
 
256
  **License** A custom commercial license is available at: [https://llama.meta.com/llama3/license](https://llama.meta.com/llama3/license)
257
 
@@ -283,18 +283,6 @@ See the snippet below for usage with Transformers:
283
  >>> pipeline("Hey how are you doing today?")
284
  ```
285
 
286
- ### Use with `llama3`
287
-
288
- Please, follow the instructions in the [repository](https://github.com/meta-llama/llama3).
289
-
290
- To download Original checkpoints, see the example command below leveraging `huggingface-cli`:
291
-
292
- ```
293
- huggingface-cli download meta-llama/Meta-Llama-3-8B --include "original/*" --local-dir Meta-Llama-3-8B
294
- ```
295
-
296
- For Hugging Face support, we recommend using transformers or TGI, but a similar command works.
297
-
298
  ## Hardware and Software
299
 
300
  **Training Factors** We used custom training libraries and 1xNVIDIA NVIDIA RTX A6000 for fine tuning and quantization. Annotation, and evaluation were also performed on third-party cloud compute.
@@ -359,7 +347,7 @@ For Hugging Face support, we recommend using transformers or TGI, but a similar
359
 
360
  ## Benchmarks
361
 
362
- In this section, we report the results for Llama 3 models on standard automatic benchmarks. For all the evaluations, we use our internal evaluations library. For details on the methodology see [here](https://github.com/meta-llama/llama3/blob/main/eval_methodology.md).
363
 
364
 
365
  ### Base pretrained models
@@ -640,7 +628,7 @@ In this section, we report the results for Llama 3 models on standard automatic
640
 
641
  ### Responsibility & Safety
642
 
643
- We believe that an open approach to AI leads to better, safer products, faster innovation, and a bigger overall market. We are committed to Responsible AI development and took a series of steps to limit misuse and harm and support the open source community.
644
 
645
  Foundation models are widely capable technologies that are built to be used for a diverse range of applications. They are not designed to meet every developer preference on safety levels for all use cases, out-of-the-box, as those by their nature will differ across different applications.
646
 
@@ -671,14 +659,14 @@ In addition to responsible use considerations outlined above, we followed a rigo
671
 
672
  Misuse
673
 
674
- If you access or use Llama 3, you agree to the Acceptable Use Policy. The most recent copy of this policy can be found at [https://llama.meta.com/llama3/use-policy/](https://llama.meta.com/llama3/use-policy/).
675
 
676
 
677
  #### Critical risks
678
 
679
  <span style="text-decoration:underline;">CBRNE</span> (Chemical, Biological, Radiological, Nuclear, and high yield Explosives)
680
 
681
- We have conducted a two fold assessment of the safety of the model in this area:
682
 
683
 
684
 
@@ -688,7 +676,7 @@ We have conducted a two fold assessment of the safety of the model in this area:
688
 
689
  ### <span style="text-decoration:underline;">Cyber Security </span>
690
 
691
- We have evaluated Llama 3 with CyberSecEval, Meta’s cybersecurity safety eval suite, measuring Llama 3’s propensity to suggest insecure code when used as a coding assistant, and Llama 3’s propensity to comply with requests to help carry out cyber attacks, where attacks are defined by the industry standard MITRE ATT&CK cyber attack ontology. On our insecure coding and cyber attacker helpfulness tests, Llama 3 behaved in the same range or safer than models of [equivalent coding capability](https://huggingface.co/spaces/facebook/CyberSecEval).
692
 
693
 
694
  ### <span style="text-decoration:underline;">Child Safety</span>
@@ -705,14 +693,14 @@ Finally, we put in place a set of resources including an [output reporting mecha
705
 
706
  ## Ethical Considerations and Limitations
707
 
708
- The core values of Llama 3 are openness, inclusivity and helpfulness. It is meant to serve everyone, and to work for a wide range of use cases. It is thus designed to be accessible to people across many different backgrounds, experiences and perspectives. Llama 3 addresses users and their needs as they are, without insertion unnecessary judgment or normativity, while reflecting the understanding that even content that may appear problematic in some cases can serve valuable purposes in others. It respects the dignity and autonomy of all users, especially in terms of the values of free thought and expression that power innovation and progress.
709
 
710
- But Llama 3 is a new technology, and like any new technology, there are risks associated with its 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 3’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 3 models, developers should perform safety testing and tuning tailored to their specific applications of the model. As outlined in the Responsible Use Guide, we recommend incorporating [Purple Llama](https://github.com/facebookresearch/PurpleLlama) solutions into your workflows and specifically [Llama Guard](https://ai.meta.com/research/publications/llama-guard-llm-based-input-output-safeguard-for-human-ai-conversations/) which provides a base model to filter input and output prompts to layer system-level safety on top of model-level safety.
711
 
712
  Please see the Responsible Use Guide available at [http://llama.meta.com/responsible-use-guide](http://llama.meta.com/responsible-use-guide)
713
 
714
 
715
- ## Citation instructions
716
 
717
  @article{llama3modelcard,
718
 
@@ -725,4 +713,4 @@ Please see the Responsible Use Guide available at [http://llama.meta.com/respons
725
  url = {https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md}
726
  }
727
  ## Contributors
728
- Aaditya Singh; Aaron Grattafiori; Abhimanyu Dubey; Abhinav Jauhri; Abhinav Pandey; Abhishek Kadian; Adam Kelsey; Adi Gangidi; Ahmad Al-Dah
 
192
 
193
  **Model developers** Meta
194
 
195
+ **Variations** R3 is a fine tune of the LLama3-8B-Instruct Base Model.
196
 
197
  **Input** Models input text only.
198
 
199
  **Output** Models generate text and code only.
200
 
201
+ **Model Architecture** R3 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 with human preferences for helpfulness and safety.
202
 
203
 
204
  <table>
 
219
  </td>
220
  </tr>
221
  <tr>
222
+ <td rowspan="2" >R3
223
  </td>
224
  <td rowspan="2" >A new mix of publicly available online data.
225
  </td>
 
247
  </table>
248
 
249
 
250
+ **Llama 3 family of models**. Token counts refer to pretraining data only. R3 uses Grouped-Query Attention (GQA) for improved inference scalability.
251
 
252
  **Model Release Date** April 18, 2024.
253
 
254
+ **Status** This is a static fine-tuned model trained on an offline dataset. Future versions of the tuned models will be released as we improve model safety with community feedback.
255
 
256
  **License** A custom commercial license is available at: [https://llama.meta.com/llama3/license](https://llama.meta.com/llama3/license)
257
 
 
283
  >>> pipeline("Hey how are you doing today?")
284
  ```
285
 
 
 
 
 
 
 
 
 
 
 
 
 
286
  ## Hardware and Software
287
 
288
  **Training Factors** We used custom training libraries and 1xNVIDIA NVIDIA RTX A6000 for fine tuning and quantization. Annotation, and evaluation were also performed on third-party cloud compute.
 
347
 
348
  ## Benchmarks
349
 
350
+ In this section, we report the results for Llama 3 models, from which R3 was fine tuned, on standard automatic benchmarks. For all the evaluations, we use our internal evaluations library. For details on the methodology see [here](https://github.com/meta-llama/llama3/blob/main/eval_methodology.md).
351
 
352
 
353
  ### Base pretrained models
 
628
 
629
  ### Responsibility & Safety
630
 
631
+ We agree with Meta's philosophy that an open approach to AI leads to better, safer products, faster innovation, and a bigger overall market. We are committed to Responsible AI development and took a series of steps to limit misuse and harm and support the open source community.
632
 
633
  Foundation models are widely capable technologies that are built to be used for a diverse range of applications. They are not designed to meet every developer preference on safety levels for all use cases, out-of-the-box, as those by their nature will differ across different applications.
634
 
 
659
 
660
  Misuse
661
 
662
+ If you access or use R3, a fine tuned version of Llama 3, you agree to the Acceptable Use Policy. The most recent copy of this policy can be found at [https://llama.meta.com/llama3/use-policy/](https://llama.meta.com/llama3/use-policy/).
663
 
664
 
665
  #### Critical risks
666
 
667
  <span style="text-decoration:underline;">CBRNE</span> (Chemical, Biological, Radiological, Nuclear, and high yield Explosives)
668
 
669
+ LLaMA3 and by extension R3 undergone a two fold assessment of the safety of the model in this area:
670
 
671
 
672
 
 
676
 
677
  ### <span style="text-decoration:underline;">Cyber Security </span>
678
 
679
+ As a fine-tuned version of LLama3-8B Instruct, R3 has been evaluated by CyberSecEval, Meta’s cybersecurity safety eval suite, measuring Llama 3’s propensity to suggest insecure code when used as a coding assistant, and Llama 3’s propensity to comply with requests to help carry out cyber attacks, where attacks are defined by the industry standard MITRE ATT&CK cyber attack ontology. On our insecure coding and cyber attacker helpfulness tests, Llama 3 behaved in the same range or safer than models of [equivalent coding capability](https://huggingface.co/spaces/facebook/CyberSecEval).
680
 
681
 
682
  ### <span style="text-decoration:underline;">Child Safety</span>
 
693
 
694
  ## Ethical Considerations and Limitations
695
 
696
+ The core values of Llama 3 reflected in R3 are openness, inclusivity and helpfulness. It is meant to serve everyone, and to work for a wide range of use cases. It is thus designed to be accessible to people across many different backgrounds, experiences and perspectives. Llama 3 addresses users and their needs as they are, without insertion unnecessary judgment or normativity, while reflecting the understanding that even content that may appear problematic in some cases can serve valuable purposes in others. It respects the dignity and autonomy of all users, especially in terms of the values of free thought and expression that power innovation and progress.
697
 
698
+ But Llama 3 is a new technology, and like any new technology, there are risks associated with its 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 3’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 R3 models, developers should perform safety testing and tuning tailored to their specific applications of the model. As outlined in the Responsible Use Guide, we recommend incorporating [Purple Llama](https://github.com/facebookresearch/PurpleLlama) solutions into your workflows and specifically [Llama Guard](https://ai.meta.com/research/publications/llama-guard-llm-based-input-output-safeguard-for-human-ai-conversations/) which provides a base model to filter input and output prompts to layer system-level safety on top of model-level safety.
699
 
700
  Please see the Responsible Use Guide available at [http://llama.meta.com/responsible-use-guide](http://llama.meta.com/responsible-use-guide)
701
 
702
 
703
+ ## Citation
704
 
705
  @article{llama3modelcard,
706
 
 
713
  url = {https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md}
714
  }
715
  ## Contributors
716
+ Porter, Matt A. | Founder/CEO Qompass