Superminaren/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-09-24 13:37:01 +02:00
Code Issues Packages Projects Releases Wiki Activity

model-conversion : add causal-compare-logits recipe (#29305)

Browse Source 
This commit adds a new recipe/target to the Makefile which allows the
logits verification to be run on pre-existing model outputs.

The motivation for this is that for large models it can take a long time
to run them models, and especially for the original model which seldom
changes this is very time consuming. With this change we can run the
original model one which will store the tokens and logits, and then
manually run the converted model and the run use this recipe to verify
them against the orignal model.
This commit is contained in:
Daniel Bevenius
2026-09-23 12:51:35 +02:00
committed by GitHub
parent 633733d0ae
commit 18f9f7bef9
1 changed files with 3 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/model-conversion/Makefile
+3
View File
@@ -68,6 +68,9 @@ causal-run-converted-model:
@CONVERTED_MODEL="$(CONVERTED_MODEL)" ./scripts/causal/run-converted-model.sh
causal-verify-logits: causal-run-original-model causal-run-converted-model
$(MAKE) causal-compare-logits
causal-compare-logits:
@MODEL_PATH="$(MODEL_PATH)" ./scripts/causal/compare-logits.py
@MODEL_PATH="$(MODEL_PATH)" ./scripts/utils/check-nmse.py -m ${MODEL_PATH}