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

Fix to skip GPU offloading so falcon models work correctly

Browse source
This commit is contained in:
Concedo 2023-08-30 18:26:41 +08:00
parent d4c22a8b02
commit 1301bd7e29
2 changed files with 13 additions and 24 deletions
Download patch file Download diff file Expand all files Collapse all files

35
expose.cpp
View file

@ -126,19 +126,6 @@ extern "C"
return true;
}
}
else if(file_format==FileFormat::RWKV_1 || file_format==FileFormat::RWKV_2)
{
printf("\n---\nIdentified as RWKV model: (ver %d)\nAttempting to Load...\n---\n", file_format);
ModelLoadResult lr = gpttype_load_model(inputs, file_format);
if (lr == ModelLoadResult::FAIL || lr == ModelLoadResult::RETRY_LOAD)
{
return false;
}
else
{
return true;
}
}
else if(file_format==FileFormat::NEOX_1 || file_format==FileFormat::NEOX_2 || file_format==FileFormat::NEOX_3 || file_format==FileFormat::NEOX_4 || file_format==FileFormat::NEOX_5 || file_format==FileFormat::NEOX_6 || file_format==FileFormat::NEOX_7)
{
printf("\n---\nIdentified as GPT-NEO-X model: (ver %d)\nAttempting to Load...\n---\n", file_format);
@ -173,22 +160,24 @@ extern "C"
return true;
}
}
else if(file_format==FileFormat::MPT_1)
else
{
printf("\n---\nIdentified as MPT model: (ver %d)\nAttempting to Load...\n---\n", file_format);
ModelLoadResult lr = gpttype_load_model(inputs, file_format);
if (lr == ModelLoadResult::FAIL || lr == ModelLoadResult::RETRY_LOAD)
if(file_format==FileFormat::MPT_1)
{
return false;
printf("\n---\nIdentified as MPT model: (ver %d)\nAttempting to Load...\n---\n", file_format);
}
else if(file_format==FileFormat::RWKV_1 || file_format==FileFormat::RWKV_2)
{
printf("\n---\nIdentified as RWKV model: (ver %d)\nAttempting to Load...\n---\n", file_format);
}
else if(file_format==FileFormat::GGUF_FALCON)
{
printf("\n---\nIdentified as FALCON model: (ver %d)\nAttempting to Load...\n---\n", file_format);
}
else
{
return true;
printf("\n---\nIdentified as LLAMA model: (ver %d)\nAttempting to Load...\n---\n", file_format);
}
}
else
{
printf("\n---\nIdentified as LLAMA model: (ver %d)\nAttempting to Load...\n---\n", file_format);
ModelLoadResult lr = gpttype_load_model(inputs, file_format);
if (lr == ModelLoadResult::FAIL || lr == ModelLoadResult::RETRY_LOAD)
{

2
klite.embd
View file

@ -6385,7 +6385,7 @@ Kobold Lite is under the AGPL v3.0 License for the purposes of koboldcpp. Please
//version 1.2.4 and later supports unban tokens
if (kobold_endpoint_version && kobold_endpoint_version != "" && compare_version_str(kobold_endpoint_version, "1.2.3") > 0)
{
submit_payload.use_default_badwordids = (localsettings.unban_tokens?true:false);
submit_payload.use_default_badwordids = (localsettings.unban_tokens?false:true);
}
let pseudostreaming = should_use_pseudostreaming();