vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

llama : sanitize invalid tokens (#9357)

Browse source 
* common : do not add null tokens during warmup

ggml-ci

* llama : check that the input tokens are valid

ggml-ci

* tests : fix batch size of bert model

ggml-ci
This commit is contained in:
Georgi Gerganov 2024-09-08 00:33:13 +03:00 • committed by GitHub
parent e536426ded
commit faf69d4237
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 26 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

14
src/llama.cpp
View file

@ -16066,6 +16066,13 @@ static int llama_decode_internal(
return -1;
}
for (uint32_t i = 0; i < n_tokens_all; ++i) {
if (batch_all.token[i] < 0) {
LLAMA_LOG_ERROR("%s: invalid token[%d] = %d", __func__, i, batch_all.token[i]);
return -1;
}
}
const auto & model = lctx.model;
const auto & hparams = model.hparams;
const auto & cparams = lctx.cparams;
@ -16358,6 +16365,13 @@ static int llama_encode_internal(
return -1;
}
for (uint32_t i = 0; i < n_tokens; ++i) {
if (batch.token[i] < 0) {
LLAMA_LOG_ERROR("%s: invalid token[%d] = %d", __func__, i, batch.token[i]);
return -1;
}
}
const auto & model = lctx.model;
const auto & hparams = model.hparams;
const auto & cparams = lctx.cparams;