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

Add Command R Plus GGUF

Browse source
This commit is contained in:
S 2024-04-04 18:23:23 +01:00
parent 0a1d889e27
commit 2efcd87b12
3 changed files with 7 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

3
convert-hf-to-gguf.py
View file

@ -149,6 +149,7 @@ class Model(ABC):
# map tensor names
new_name = tensor_map.get_name(name, try_suffixes=(".weight", ".bias"))
if new_name is None:
print(tensor_map)
print(f"Can not map tensor {name!r}")
sys.exit()
@ -2344,6 +2345,8 @@ class CommandR2Model(Model):
# max_position_embeddings = 8192 in config.json but model was actually
# trained on 128k context length
if "model_max_length" not in self.hparams:
self.hparams["model_max_length"] = 131072
self.hparams["max_position_embeddings"] = self.hparams["model_max_length"]
def set_gguf_parameters(self):

2
gguf-py/gguf/constants.py
View file

@ -638,6 +638,8 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
MODEL_TENSOR.ATTN_K_NORM,
MODEL_TENSOR.ATTN_Q_NORM,
],
# TODO
}

2
gguf-py/gguf/tensor_mapping.py
View file

@ -285,12 +285,14 @@ class TensorNameMap:
MODEL_TENSOR.ATTN_Q_NORM: (
"language_model.encoder.layers.{bid}.self_attention.q_layernorm",
"model.layers.{bid}.self_attn.q_layernorm", # persimmon
"model.layers.{bid}.self_attn.q_norm", # cohere
"transformer.blocks.{bid}.attn.q_ln", # sea-lion
),
MODEL_TENSOR.ATTN_K_NORM: (
"language_model.encoder.layers.{bid}.self_attention.k_layernorm",
"model.layers.{bid}.self_attn.k_layernorm", # persimmon
"model.layers.{bid}.self_attn.k_norm", # cohere
"transformer.blocks.{bid}.attn.k_ln", # sea-lion
),