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

handle stablecode-completion-alpha-3b

Browse source
This commit is contained in:
Concedo 2023-08-09 21:07:57 +08:00
parent 90058d96b0
commit 3a7853d259
1 changed files with 26 additions and 18 deletions
Download patch file Download diff file Expand all files Collapse all files

16
model_adapter.cpp
View file

@ -133,11 +133,18 @@ void print_tok_vec(std::vector<float> &embd)
else if(vocabsiz==50257 || (vocabsiz>=49152&&vocabsiz<=49157)) //49152-6 is starcoder else if(vocabsiz==50257 || (vocabsiz>=49152&&vocabsiz<=49157)) //49152-6 is starcoder
{ {
fileformat = FileFormat::GPT2_1; fileformat = FileFormat::GPT2_1;
uint32_t temp; uint32_t temp, v1,v2,v3;
fin.read((char *)&temp, sizeof(temp)); //ctx fin.read((char *)&v1, sizeof(temp)); //ctx
fin.read((char *)&temp, sizeof(temp)); //n_embd fin.read((char *)&v2, sizeof(temp)); //n_embd
fin.read((char *)&temp, sizeof(temp)); //n_head fin.read((char *)&v3, sizeof(temp)); //n_head
fin.read((char *)&temp, sizeof(temp)); //n_layer fin.read((char *)&temp, sizeof(temp)); //n_layer
if(vocabsiz==49152 && v1==4096 && v2==2560 && v3==32 && temp==32)
{
//special case, Stablecode Completion Alpha 3B
fileformat = FileFormat::NEOX_6;
}
else
{
fin.read((char *)&temp, sizeof(temp)); //f16 fin.read((char *)&temp, sizeof(temp)); //f16
const int32_t qntvr = temp / 1000; const int32_t qntvr = temp / 1000;
temp %= 1000; temp %= 1000;
@ -157,6 +164,7 @@ void print_tok_vec(std::vector<float> &embd)
fileformat = FileFormat::GPT2_2; //quantized format cannot be legacy type fileformat = FileFormat::GPT2_2; //quantized format cannot be legacy type
} }
} }
}
else if(vocabsiz < 31998 || vocabsiz > 33000) else if(vocabsiz < 31998 || vocabsiz > 33000)
{ {
//anything outside the llama v1 range is assumed to be NeoX //anything outside the llama v1 range is assumed to be NeoX