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

convert-hf-to-gguf.py: print --> logger.debug or ValueError()

Browse source
This commit is contained in:
brian khuu 2024-04-18 14:26:36 +10:00
parent 3a55ae4d72
commit aefd7492a3
1 changed files with 9 additions and 13 deletions
Download patch file Download diff file Expand all files Collapse all files

22
convert-hf-to-gguf.py
View file

@ -1323,8 +1323,7 @@ class StableLMModel(Model):
# map tensor names # map tensor names
new_name = tensor_map.get_name(name, try_suffixes=(".weight", ".bias")) new_name = tensor_map.get_name(name, try_suffixes=(".weight", ".bias"))
if new_name is None: if new_name is None:
print(f"Can not map tensor {name!r}") raise ValueError(f"Can not map tensor {name!r}")
sys.exit()
n_dims = len(data.shape) n_dims = len(data.shape)
data_dtype = data.dtype data_dtype = data.dtype
@ -1341,7 +1340,7 @@ class StableLMModel(Model):
if self.ftype == 1 and data_dtype == np.float32 and name.endswith(".weight") and not new_name.endswith("_norm.weight") and n_dims == 2: if self.ftype == 1 and data_dtype == np.float32 and name.endswith(".weight") and not new_name.endswith("_norm.weight") and n_dims == 2:
data = data.astype(np.float16) data = data.astype(np.float16)
print(f"{new_name}, n_dims = {n_dims}, {old_dtype} --> {data.dtype}") logger.debug(f"{new_name}, n_dims = {n_dims}, {old_dtype} --> {data.dtype}")
self.gguf_writer.add_tensor(new_name, data) self.gguf_writer.add_tensor(new_name, data)
@ -1357,8 +1356,7 @@ class StableLMModel(Model):
merged_name = f"model.layers.{bid}.self_attn.{layer_name}.weight" merged_name = f"model.layers.{bid}.self_attn.{layer_name}.weight"
new_name = tensor_map.get_name(merged_name, try_suffixes=(".weight", ".bias")) new_name = tensor_map.get_name(merged_name, try_suffixes=(".weight", ".bias"))
if new_name is None: if new_name is None:
print(f"Can not map tensor {name!r}") raise ValueError(f"Can not map tensor {name!r}")
sys.exit()
if self.ftype == 1 and data_dtype == np.float16 and (n_dims == 1 or new_name.endswith("_norm.weight")): if self.ftype == 1 and data_dtype == np.float16 and (n_dims == 1 or new_name.endswith("_norm.weight")):
data = data.astype(np.float32) data = data.astype(np.float32)
@ -1366,7 +1364,7 @@ class StableLMModel(Model):
if self.ftype == 1 and data_dtype == np.float32 and name.endswith(".weight") and not new_name.endswith("_norm.weight") and n_dims == 2: if self.ftype == 1 and data_dtype == np.float32 and name.endswith(".weight") and not new_name.endswith("_norm.weight") and n_dims == 2:
data = data.astype(np.float16) data = data.astype(np.float16)
print(f"{new_name}, n_dims = {len(data.shape)}, shape = {data.shape} --> {data.dtype}") logger.debug(f"{new_name}, n_dims = {len(data.shape)}, shape = {data.shape} --> {data.dtype}")
self.gguf_writer.add_tensor(new_name, data) self.gguf_writer.add_tensor(new_name, data)
@ -1933,10 +1931,9 @@ class Qwen2MoeModel(Model):
new_name = tensor_map.get_name(merged_name, try_suffixes=(".weight", ".bias")) new_name = tensor_map.get_name(merged_name, try_suffixes=(".weight", ".bias"))
if new_name is None: if new_name is None:
print(f"Can not map tensor {name!r}") raise ValueError(f"Can not map tensor {name!r}")
sys.exit()
print(f"{new_name}, n_dims = {len(data.shape)}, shape = {data.shape} --> {data.dtype}") logger.debug(f"{new_name}, n_dims = {len(data.shape)}, shape = {data.shape} --> {data.dtype}")
self.gguf_writer.add_tensor(new_name, data) self.gguf_writer.add_tensor(new_name, data)
continue continue
@ -1944,8 +1941,7 @@ class Qwen2MoeModel(Model):
# map tensor names # map tensor names
new_name = tensor_map.get_name(name, try_suffixes=(".weight", ".bias")) new_name = tensor_map.get_name(name, try_suffixes=(".weight", ".bias"))
if new_name is None: if new_name is None:
print(f"Can not map tensor {name!r}") raise ValueError(f"Can not map tensor {name!r}")
sys.exit()
n_dims = len(data.shape) n_dims = len(data.shape)
data_dtype = data.dtype data_dtype = data.dtype
@ -1962,7 +1958,7 @@ class Qwen2MoeModel(Model):
if self.ftype == 1 and data_dtype == np.float32 and name.endswith(".weight") and n_dims == 2: if self.ftype == 1 and data_dtype == np.float32 and name.endswith(".weight") and n_dims == 2:
data = data.astype(np.float16) data = data.astype(np.float16)
print(f"{new_name}, n_dims = {n_dims}, shape = {data.shape}, {old_dtype} --> {data.dtype}") logger.debug(f"{new_name}, n_dims = {n_dims}, shape = {data.shape}, {old_dtype} --> {data.dtype}")
self.gguf_writer.add_tensor(new_name, data) self.gguf_writer.add_tensor(new_name, data)
@ -2642,7 +2638,7 @@ class GemmaModel(Model):
# lm_head is not used in llama.cpp, while autoawq will include this tensor in model # lm_head is not used in llama.cpp, while autoawq will include this tensor in model
# To prevent errors, skip loading lm_head.weight. # To prevent errors, skip loading lm_head.weight.
if name == "lm_head.weight": if name == "lm_head.weight":
print(f"Skipping get tensor {name!r} in safetensors so that convert can end normally.") logger.debug(f"Skipping get tensor {name!r} in safetensors so that convert can end normally.")
continue continue
old_dtype = data_torch.dtype old_dtype = data_torch.dtype