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

fix: revert some changes

Browse source
This commit is contained in:
Joan Martinez 2024-04-22 13:20:03 +02:00
parent e2323706e4
commit 795ff1d3d3
2 changed files with 5 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

5
convert-hf-to-gguf.py
View file

@ -92,7 +92,6 @@ class Model(ABC):
def set_gguf_parameters(self):
self.gguf_writer.add_name(self.dir_model.name)
print(f'self.block_count {self.block_count}')
self.gguf_writer.add_block_count(self.block_count)
if (n_ctx := self.find_hparam(["max_position_embeddings", "n_ctx"], optional=True)) is not None:
@ -138,7 +137,6 @@ class Model(ABC):
def write_tensors(self):
block_count = self.hparams.get("n_layers", self.hparams.get("num_hidden_layers", self.hparams.get("n_layer")))
tensor_map = gguf.get_tensor_name_map(self.model_arch, block_count)
print(f'Block_count {block_count} with tensor_map {tensor_map}')
for name, data_torch in self.get_tensors():
# we don't need these
if name.endswith((".attention.masked_bias", ".attention.bias", ".attention.rotary_emb.inv_freq")):
@ -2188,6 +2186,9 @@ class JinaBertModel(BertModel):
yield name, data
JinaBertForMaskedML = JinaBertModel
@Model.register("GemmaForCausalLM")
class GemmaModel(Model):
model_arch = gguf.MODEL_ARCH.GEMMA

3
ggml.c
View file

@ -12476,8 +12476,9 @@ static void ggml_compute_forward_alibi_f32(
for (int64_t i = 0; i < ne0; i++) {
for (int64_t j = 0; j < ne1; j++) {
float * const src = (float *)((char *) src0->data + i*nb0 + j*nb1 + k*nb2);
float * pdst = (float *)((char *) dst->data + i*nb0 + j*nb1 + k*nb2);
pdst[0] = -1.0f * i * m_k;
pdst[0] = i * m_k + src[0];
}
}
}