thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 4

chore : correct typos [no ci] (#20041)

Browse source 
* fix(docs): correct typos found during code review

Non-functional changes only:
- Fixed minor spelling mistakes in comments
- Corrected typos in user-facing strings
- No variables, logic, or functional code was modified.

Signed-off-by: Marcel Petrick <mail@marcelpetrick.it>

* Update docs/backend/CANN.md

Co-authored-by: Aaron Teo <taronaeo@gmail.com>

* Revert "Auxiliary commit to revert individual files from 846d1c301281178efbc6ce6060ad34c1ebe45af8"

This reverts commit 02fcf0c7db661d5ff3eff96b2b2db9fdb7213256.

* Update tests/test-backend-ops.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

* Update tests/test-backend-ops.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

---------

Signed-off-by: Marcel Petrick <mail@marcelpetrick.it>
Co-authored-by: Aaron Teo <taronaeo@gmail.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
This commit is contained in:
Marcel Petrick 2026-03-05 08:50:21 +01:00 • committed by GitHub
parent 7a99dc85e2
commit 92f7da00b4
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
81 changed files with 160 additions and 160 deletions
Download patch file Download diff file Expand all files Collapse all files

4
tools/perplexity/README.md
View file

@ -27,10 +27,10 @@ In addition to the KL divergence the following statistics are calculated with `-
* Ratio of mean FP16 PPL and quantized PPL. Uncertainty is estimated on logits, then propagated. The logarithm of this metric is also calculated and printed, it is 0 if the logit distributions are the same.
* Difference of mean FP16 PPL and quantized PPL. Uncertainty is estimated on logits, then propagated.
* Mean change in "correct" token probability. Positive values mean the model gets better at prediction, negative values mean it gets worse.
* Pearson correlation coefficient of the "correct" token probabilites between models.
* Pearson correlation coefficient of the "correct" token probabilities between models.
* Percentiles of change in "correct" token probability. Positive values mean the model gets better at prediction, negative values mean it gets worse. Can be used to judge noise vs. quality loss from quantization. If the percentiles are symmetric then the quantization is essentially just adding noise. If the negative values are significantly larger than the positive values then this indicates that the model is actually becoming worse from the quantization.
* The root mean square of the change in token probabilities. If you were to assume that the quantization simply causes Gaussian noise on the token probabilities then this would be the standard deviation of said noise. The uncertainty on the value is calculated that the change in token probabilities follows a Gaussian distribution. Related discussion: https://github.com/ggml-org/llama.cpp/discussions/2875 .
* Same top p: Percentage of how often the token was assigned the highest probabilites by both models. The uncertainty is calculated from the Gaussian approximation of the binomial distribution.
* Same top p: Percentage of how often the token was assigned the highest probabilities by both models. The uncertainty is calculated from the Gaussian approximation of the binomial distribution.
## LLaMA 3 8b Scoreboard