forked from ggml-org/llama.cpp
-
-
Notifications
You must be signed in to change notification settings - Fork 0
Commit a18f481
server : use common_token_to_piece instead of common_detokenize (ggml-org#11740)
* server : use common_token_to_piece instead of common_detokenize
This commit replaces the call to common_detokenize with
common_token_to_piece in the populate_token_probs.
The motivation for this change is to avoid an issue where
common_detokenize would remove the word boundary character for tokens,
which caused a regression in the server generated token probabilities.
Resolves: ggml-org#11728
* squash! server : use common_token_to_piece instead of common_detokenize
Use common_token_to_piece for post_sampling_probs as well.1 parent b9ab0a4 commit a18f481
1 file changed
+2
-2
lines changed| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
2279 | 2279 | | |
2280 | 2280 | | |
2281 | 2281 | | |
2282 | - | ||
2282 | + | ||
2283 | 2283 | | |
2284 | 2284 | | |
2285 | 2285 | | |
| |||
2301 | 2301 | | |
2302 | 2302 | | |
2303 | 2303 | | |
2304 | - | ||
2304 | + | ||
2305 | 2305 | | |
2306 | 2306 | | |
2307 | 2307 | | |
| |||
0 commit comments