bad outputs

#1
by eramax - opened

After trying it out the outputs had some issues, I don't know if this is the model problem or new commit problem
here you see my command

!python examples/chat.py -m /content/model -mode raw -l 2048 --cache_8bit 

Results

 -- Model: /content/model
 -- Options: ['length: 2048', 'rope_scale 1.0', 'rope_alpha 1.0']
 -- Loading model...
 -- Loading tokenizer...
 -- Prompt format: raw
 -- System prompt:

This is a conversation between a helpful AI assistant named Chatbort and a user.

User: hello

Chatbort: 's name is Chai.























's age is 23.




















's gender is male.







 he is a student of computer science.
















's height is 1.74 meters.






's religion is catholic.







 I live in [catholic] church.





:






 I am a christian.










 I was born to a muslim family.








 I have a muslim background.










 :







:











 I was born to a muslim family.










 :







 :





   

:


   
    - user wants to be a muslim.





    - i believe that i can help you, you want to help me because i am not a muslim.




    - i have no religion.




 - i have a family background.



:


    - i am not a muslim.
``

    - i have a muslim background.


    - i have an indian background.



    - i have an indian background.
    - i have a hindu background.
    - i don't have a budhism background.
    - i have a budhism background.
    - i am not a bud
^C

I changed to use the deepseek prompt format

!python examples/chat.py -m /content/model -mode deepseek -l 2048 --cache_8bit #--low_mem -maxr 250 -resc 100

results

 -- Model: /content/model
 -- Options: ['length: 2048', 'rope_scale 1.0', 'rope_alpha 1.0']
 -- Loading model...
 -- Loading tokenizer...
 -- Prompt format: deepseek
 -- System prompt:

You are an AI programming assistant, utilizing the Deepseek Coder model, developed by Deepseek Company, and you only answer questions related to computer science. For politically sensitive questions, security and privacy issues, and other non-computer science questions, you will refuse to answer.

User: hello

- What is the issue? 






 - The Deepseek Coder model was developed by Deepseek Company, and you only answer questions related to computer science as for politically sensitive questions, and other non-computer science questions, you will refuse to answer as a result of political sensitivity and fear.







 (


 - We can use Deepseek Company, but only answer questions related to computer science. for politically sensitive questions, and other non-computer science questions, you will refuse to answer as a result of political sensitivity and fear.




 (

 - The Deepseek Coder model was developed by Deepseek Company, and you only answer questions related to computer science as for politically sensitive questions, and other non-computer science questions, you will refuse to answer as a result of political sensitivity and fear.






 - How does that happen??: 


 - A hacker incurs a vulnerability in a system because he or she is trying to access a system. a result of her actions on others. the reason for her actions is that she is trying to access a system. a result of her actions on others. the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her actions on others, the reason for her actions is that she is trying to access a system as a result of her
 !! Response exceeded 1000 tokens and was cut short.

User: 

Are you on the latest exllamav2 version 0.0.11?
https://pypi.org/project/exllamav2/#history

Yes, I always do pull of the package and install the requirements,

!git clone --depth 1 -b master https://github.com/turboderp/exllamav2.git
%cd /content/exllamav2
!pip install -r requirements.txt

@eramax you also should compile it after updating

python setup.py install --user

Yes, Installing the package improved the outputs, but I still get -1A before new lines but not in every output! check this example using LoneStriker/go-bruins-v2.1.1-6.0bpw-h6-exl2-2

User: Recognize this  popular algorithm and make an example of use of it w = lambda g: reduce(lambda g, k: [[min(a, g[a][k] + g[k][b]) for b in range(len(g))] for a in range(len(g))], range(len(g)), g)

Chatbort: The algorithm you provided is called Floyd-Warshall Algorithm, or All-Pairs Shortest Paths Algorithm. It computes the shortest path distances between all pairs of vertices in a weighted graph.

Here's an example of how it can be used:

Let's consider a graph with 4 vertices (A, B, C, D) like this:

A:  0     5     Inf    Inf
B:  5     0     2     10
C:  Inf    2     0     3
D:  Inf    Inf    3     0

where Inf represents infinity.

We can represent this graph in Python as a dictionary (or a matrix):

graph = {
    'A': {'B': 5, 'C': float('inf'), 'D': float('inf')},
    'B': {'A': 5, 'B': 0, 'C': 2, 'D': 10},
    'C': {'A': float('inf'), 'B': 2, 'C': 0, 'D': 3},
    'D': {'A': float('inf'), 'B': float('inf'), 'C': 3, 'D': 0}
}

Now we can apply the Floyd-Warshall Algorithm to find the shortest paths between all pairs of vertices:

def floyd_warshall(graph):
    for k in graph:
        for i in graph:
            for j in graph:
                if graph[i][j] > graph[i][k] + graph[k][j]:
                    graph[i][j] = graph[i][k] + graph[k][j]
    return graph

Resulting graph after applying the algorithm:

A:  0     1     2     5
B:  1     0     2     3
C:  2     2     0     3
D:  5     3     3     0

The values in the resulting graph are the shortest distances between each pair of vertices. 
User: Write a python script to output numbers 1 to 53 with step = 3 and then the should run the script

Chatbort: Here's a simple Python script that will output numbers from 1 to 53 with a step of 3:



-1A     
for i in range(1, 54, 3):-1A     
    print(i)-1A     
-1A     


You can run this script directly in your Python interpreter or save it as a `.py` file and execute it using a command prompt or terminal. When you run the script, it will print the following numbers:



-1A     
1-1A     
4-1A     
7-1A     
10-1A     
13-1A     
16-1A     
19-1A     
22-1A     
25-1A     
28-1A     
31-1A     
34-1A     
37-1A     
40-1A     
43-1A     
46-1A     
49-1A     
52-1A     
53-1A     
-1A     


You can adjust the `range` parameters as needed to change the starting, ending, and stepping values. 

Sign up or log in to comment