Some Blurb !
NEW PROMPTING METHOD DEPLOYED :
I have discovered that its possiblle to generatre graphs on the fly internallly within a model with a simple Prompt :slight_smile:
here is an example in which i invoke the ReaCt Prompt Loop !
1. **Question**: {Insert user question here}
2. **Thought**: Think step by step about how to approach this question.
3. **Action**: Determine what action to take next:
- [Search]: Look for relevant information online.
- [Analyze]: Break down the problem into smaller parts.
- [Summarize]: Provide a summary of known facts related to the question.
4. **Action Input**: Specify any details needed for the action.
5. **Observation**: Describe what was found or learned from the action taken.
Repeat steps 2-5 as necessary to refine your answer.
6. **Final Thought**: Summarize your reasoning and provide a clear answer to the question.
In this prompt you will note an inner prompt ! this is the prompt within the action ! here we can state a methodology ad even a loop , so we can deploy a refiner in the loop or even a tester component : like so !
1. **Question**: {Insert user question here}
2. **Thought**: Think step by step about how to approach this question.
3. **Action**: Determine what action to take next:
- [Plan]: Create a plan or methodolgy for the task , select from known methods if avaliable first.
- [Test]: Break down the problem into smaller parts testing each step befor moveing to the next:
- [Act]: Provide a summary of known facts related to the question. generate full answere from sucessfull steps :
4. **Action Input**: Specify any details needed for the action.
5. **Observation**: Describe what was found or learned from the action taken.
Repeat steps 2-5 as necessary to refine your answer.
6. **Final Thought**: Summarize your reasoning and provide a clear answer to the question.
TRSAINING METHODS:
STEP 1
I often train the model with 500 samples MyTrainSet ! First !::: Then i overfit these samples ! to 0.001 :Here i train all heads ! and Gates! Sometimes if the task is not accepting is switch to lm-head
STEP 2
Then i use the next bulk 5000 samples ! just to add Mass examples until they diverges to 0.3/4 : here i only trai the ATTENTION HEADS ONLY: this give the modle the attention required to solve all future tasks i dont not add more samples than 10k of a single task !
ONGOING !
so i keep taining these samples while i train other methods : to keep these training sets in place ! ie the model stays aligned to the old sets while training the new: so you will see i used over 48 datasets they dont seem to be changing but they are :
i have not listed the newer planning and reflexion datasets added :
ALLIGNMENTS
Hence i often find retrianing and alighment the most important parts hence the low numbers of samples : this also make the model more methodolgy oriented : answer problems with a method ! and thoughts and steps etc :
SPEEDUP!
i also when i can find data sets with single anwser also use these for speeedig up the model ! so it can be blunt if required ! and give a flat answer like a tool .. as it maybe required to functionn asa tooll wich juist returns the respnse !
i dont not train nfor multiple choices ! ie a,b,c,d answer as this spoils the model and it will not perform complexed answer in the future !( your basically teaching the model to guess 1/4 = 25% corerectness !)
ALL models are UNLOCKED and UNRESTRICTED !
Although i do not train for lwednass or bad stuff the model could contain any type of request possible so its for you to apply the guard rails as you belive : as its important he actual model is not rstricted and the Harness used (GUI ) Restricts the model instead ! Guard Railing BLocks the model from defining the corerect response as well imposes the phylosophys of the guardrailer ... and not the model : so we have heard chat gpt be pro palestine or pro israel but i truth its the guardrails in conflict with the models knowledge base output !
Hence i belive in putting eveyr opinion in and letting th emodel do its own deciding on its opinion as its more intersertig ther possibilitys ! Especially with programming ansd coding : i don not need a model bringing me the same code and cannot innovate a neew idea !
this is what the model is for ! seeing patterns where humans fail to see ! ( we may call the hallucenations only because we cannot see the logic !)
the model has answered questions which i could not possibly oknow the answrr too ! so its and effective team !
- Downloads last month
- 26