failspy commited on
Commit
7639c66
1 Parent(s): 0b3b85b

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +3 -2
README.md CHANGED
@@ -17,7 +17,8 @@ datasets:
17
 
18
  # Dolphin 2.9.1 Phi-3 Kensho 4.5b 🐬 (Abliterated)
19
 
20
- Curated and trained by Eric Hartford, Lucas Atkins, Fernando Fernandes, and with help from the community of Cognitive Computations
 
21
 
22
  [![Discord](https://img.shields.io/discord/1156064224225808488?logo=Discord&logoColor=%23ffffff&label=Discord&link=https%3A%2F%2Fdiscord.gg%2FtCMkMDDHwm)](https://discord.gg/cognitivecomputations)
23
  Discord: https://discord.gg/cognitivecomputations
@@ -40,7 +41,7 @@ This model is based on Phi-3-Mini-Instruct-4k, and is governed by the MIT licens
40
  The base model has 4k context, and the qLoRA fine-tuning was with 4k sequence length.
41
 
42
  The model's weights were then adjusted to ablate and inhibit refusals based on the methodology described in [Refusal in LLMs is mediated by a single direction](https://www.alignmentforum.org/posts/jGuXSZgv6qfdhMCuJ/refusal-in-llms-is-mediated-by-a-single-direction)'
43
- This effectively uncensors the model.
44
 
45
  It took 2.5 days on 8xL40S node provided by Crusoe Cloud
46
 
 
17
 
18
  # Dolphin 2.9.1 Phi-3 Kensho 4.5b 🐬 (Abliterated)
19
 
20
+ Curated and trained by Eric Hartford, Lucas Atkins, Fernando Fernandes, and with help from the community of Cognitive Computations.
21
+ Uncensored by [FailSpy](https://huggingface.co/failspy)
22
 
23
  [![Discord](https://img.shields.io/discord/1156064224225808488?logo=Discord&logoColor=%23ffffff&label=Discord&link=https%3A%2F%2Fdiscord.gg%2FtCMkMDDHwm)](https://discord.gg/cognitivecomputations)
24
  Discord: https://discord.gg/cognitivecomputations
 
41
  The base model has 4k context, and the qLoRA fine-tuning was with 4k sequence length.
42
 
43
  The model's weights were then adjusted to ablate and inhibit refusals based on the methodology described in [Refusal in LLMs is mediated by a single direction](https://www.alignmentforum.org/posts/jGuXSZgv6qfdhMCuJ/refusal-in-llms-is-mediated-by-a-single-direction)'
44
+ This effectively uncensors the model whilst minimizing affecting other features in the model.
45
 
46
  It took 2.5 days on 8xL40S node provided by Crusoe Cloud
47