Karko commited on
Commit
e49510d
·
verified ·
1 Parent(s): f59088f

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +7 -4
README.md CHANGED
@@ -7,15 +7,18 @@ tags:
7
  ---
8
  ![img_text](./assets/tmpd2xdo_x4.png)
9
 
 
10
  Proctora is a MoE model made of
11
  - OpenPipe/mistral-ft-optimized-1227 as a base model
12
  - SanjiWatsuki/Kunoichi-7B as a first expert dedicated to RP tasks.
13
  - samir-fama/SamirGPT-v1 as a second expert for factual answers.
14
 
15
- I do not have yet any metrics for this model but subjective ones. It was made at first out of curiosity and experimentation.
 
 
16
 
17
- My goal is to produce a model excellent at being a game master for RPG sessions. However being dissatisfied with the existing evaluation tool-sets, I decided to create my own (still a WIP on 01/16/24). And among my collection of models of small/medium models Proctora gave me the best results to evaluate the answers produced by other LLMs. Therefore, I surprisingly settled with it and gave it an appropriate name according to the task.
18
 
19
- TLDR: Proctora is a tool for a tool!
20
 
21
- I doubt this model will be useful for the community. I publish it for the sake of transparency in my creative process.
 
7
  ---
8
  ![img_text](./assets/tmpd2xdo_x4.png)
9
 
10
+
11
  Proctora is a MoE model made of
12
  - OpenPipe/mistral-ft-optimized-1227 as a base model
13
  - SanjiWatsuki/Kunoichi-7B as a first expert dedicated to RP tasks.
14
  - samir-fama/SamirGPT-v1 as a second expert for factual answers.
15
 
16
+ Being based on Mixtral architecture it has a natural context length of 32K, which is great.
17
+
18
+ On Openllm leaderboard it achieves a score of 71.88 which is interesting to some extent but does not really reflect the intented capacities of the model.
19
 
20
+ This model has been originally produced as a result of experimentations with mergekit. Then among my collection of LLMs, Proctora has been selected to be the "grader" in an AI-RPG evaluation suite that I am currently building. Indeed, it produced the intended grades according to given rubrics more often than other "higher performing" models in the leaderboard.
21
 
22
+ However, I also tested it in various RP scenarii using text-generation-webui (putting the character card in the system parameters and/or other world information), and I was quite impressed by the quality of the logic (relatively to other popular RP models). For example, it took in account special powers limitations better than other models. Or it managed curse activations and weaknesses better than other models that are about twice the size. Also when acting as the player (and the user being the game master), Proctora was not only able to play in character but also sometimes to make clever decision to achieve its objectives.
23
 
24
+ Having the excellent SanjiWatsuki/Kunoichi-7B as an expert, the model is uncensored. Use with caution.