mirror of
https://github.com/LostRuins/koboldcpp.git
synced 2026-09-17 08:19:49 +02:00
Dynamic sizes for sequences (#1157)
* Dynamic sizes for sequences * cleanup PR - move all dynamic fields to end of payload, ensure correct null handling to match existing behavior, add anti abuse limit of max 512 for dynamic fields * adjust anti abuse limits --------- Co-authored-by: Concedo <39025047+LostRuins@users.noreply.github.com>
This commit is contained in:
+4
-4
@@ -2488,7 +2488,7 @@ generation_outputs gpttype_generate(const generation_inputs inputs)
|
||||
double time0 = 0, time1 = 0, time2 = 0;
|
||||
timer_start();
|
||||
|
||||
for(int x=0;x<stop_token_max;++x)
|
||||
for(int x=0;x<inputs.stop_sequence_len;++x)
|
||||
{
|
||||
std::string stopper = inputs.stop_sequence[x];
|
||||
if(stopper!="")
|
||||
@@ -2516,7 +2516,7 @@ generation_outputs gpttype_generate(const generation_inputs inputs)
|
||||
delayed_generated_tokens_limit = 0;
|
||||
antislop_banned_token_ids.clear();
|
||||
banned_tokens.clear();
|
||||
for(int x=0;x<ban_token_max;++x)
|
||||
for(int x=0;x<inputs.banned_tokens_len;++x)
|
||||
{
|
||||
std::string word = inputs.banned_tokens[x];
|
||||
word = toLowerCase(word);
|
||||
@@ -2574,7 +2574,7 @@ generation_outputs gpttype_generate(const generation_inputs inputs)
|
||||
}
|
||||
|
||||
logit_biases.clear();
|
||||
for(int x=0;x<logit_bias_max;++x)
|
||||
for(int x=0;x<inputs.logit_biases_len;++x)
|
||||
{
|
||||
int32_t t_id = inputs.logit_biases[x].token_id;
|
||||
float bias = inputs.logit_biases[x].bias;
|
||||
@@ -2652,7 +2652,7 @@ generation_outputs gpttype_generate(const generation_inputs inputs)
|
||||
|
||||
if (kcpp_data->dry_multiplier > 0)
|
||||
{
|
||||
for (int x = 0; x < dry_seq_break_max; ++x)
|
||||
for (int x = 0; x < inputs.dry_sequence_breakers_len; ++x)
|
||||
{
|
||||
std::string word = inputs.dry_sequence_breakers[x];
|
||||
if (word != "")
|
||||
|
||||
Reference in New Issue
Block a user