mradermacher/Darkest-muse-v1-GGUF
Updated
•
1.1k
•
3
Pablo Dias
Anderson452
AI & ML interests
None yet
Recent Activity
liked
a model
2 months ago
mradermacher/Darkest-muse-v1-GGUF
liked
a model
2 months ago
sam-paech/Darkest-muse-v1
Organizations
None yet
Anderson452's activity
liked
2
models
2 months ago
Thank you very much, this model is the best
1
#3 opened 3 months ago
by
Anderson452
Please, 8B version
6
#8 opened 4 months ago
by
Anderson452
I ask you for a huge favor
1
#5 opened 4 months ago
by
Anderson452
liked
5
Spaces
4 months ago
reacted to
bartowski's
post with ❤️
4 months ago
Post
10046
So turns out I've been spreading a bit of misinformation when it comes to imatrix in llama.cpp
It starts true; imatrix runs the model against a corpus of text and tracks the activation of weights to determine which are most important
However what the quantization then does with that information is where I was wrong.
I think I made the accidental connection between imatrix and exllamav2's measuring, where ExLlamaV2 decides how many bits to assign to which weight depending on the goal BPW
Instead, what llama.cpp with imatrix does is it attempts to select a scale for a quantization block that most accurately returns the important weights to their original values, ie minimizing the dequantization error based on the importance of activations
The mildly surprising part is that it actually just does a relatively brute force search, it picks a bunch of scales and tries each and sees which one results in the minimum error for weights deemed important in the group
But yeah, turns out, the quantization scheme is always the same, it's just that the scaling has a bit more logic to it when you use imatrix
Huge shoutout to @compilade for helping me wrap my head around it - feel free to add/correct as well if I've messed something up
It starts true; imatrix runs the model against a corpus of text and tracks the activation of weights to determine which are most important
However what the quantization then does with that information is where I was wrong.
I think I made the accidental connection between imatrix and exllamav2's measuring, where ExLlamaV2 decides how many bits to assign to which weight depending on the goal BPW
Instead, what llama.cpp with imatrix does is it attempts to select a scale for a quantization block that most accurately returns the important weights to their original values, ie minimizing the dequantization error based on the importance of activations
The mildly surprising part is that it actually just does a relatively brute force search, it picks a bunch of scales and tries each and sees which one results in the minimum error for weights deemed important in the group
But yeah, turns out, the quantization scheme is always the same, it's just that the scaling has a bit more logic to it when you use imatrix
Huge shoutout to @compilade for helping me wrap my head around it - feel free to add/correct as well if I've messed something up
- 5 replies
updated
a
Space
5 months ago
So hear me out...
1
#7 opened 6 months ago
by
InvictusCreations
Please, add GGUF version!
1
#2 opened 7 months ago
by
Anderson452
liked
a
model
8 months ago
Great Model!!
2
#4 opened 8 months ago
by
Anderson452
Great Model!
#1 opened 8 months ago
by
Anderson452
It's an excellent version, but...
20
#1 opened 8 months ago
by
Anderson452
Error
#1 opened 8 months ago
by
Anderson452
Thank you so much!
4
#1 opened 8 months ago
by
Anderson452