mirror of
https://github.com/ggerganov/whisper.cpp.git
synced 2025-01-13 17:38:36 +01:00
models : minor changes to the HF convert script (#157)
This commit is contained in:
parent
93482d0373
commit
d64d6ca3fd
@ -28,6 +28,7 @@ conv_map = {'self_attn_layer_norm': 'attn_ln',
|
||||
'decoder.layer_norm.weight': 'decoder.ln.weight',
|
||||
'decoder.embed_positions.weight': 'decoder.positional_embedding',
|
||||
'decoder.embed_tokens.weight': 'decoder.token_embedding.weight',
|
||||
'proj_out.weight': 'decoder.proj.weight',
|
||||
}
|
||||
|
||||
# ref: https://github.com/openai/gpt-2/blob/master/src/encoder.py
|
||||
@ -82,8 +83,11 @@ fname_out = dir_out + "/ggml-model.bin"
|
||||
with open(dir_tokenizer + "/vocab.json", "r", encoding="utf8") as f:
|
||||
tokens = json.load(f)
|
||||
|
||||
|
||||
# use 16-bit or 32-bit floats
|
||||
use_f16 = True
|
||||
if len(sys.argv) > 4:
|
||||
use_f16 = False
|
||||
fname_out = dir_out + "/ggml-model-f32.bin"
|
||||
|
||||
fout = open(fname_out, "wb")
|
||||
|
||||
@ -119,6 +123,8 @@ for key in tokens:
|
||||
|
||||
list_vars = model.state_dict()
|
||||
for name in list_vars.keys():
|
||||
# this seems to not be used
|
||||
# ref: https://github.com/huggingface/transformers/blob/9a5b84a0076a04fe9596da72e8668069d4f09ea0/src/transformers/models/whisper/modeling_whisper.py#L1099-L1106
|
||||
if name == "proj_out.weight":
|
||||
print('Skipping', name)
|
||||
continue
|
||||
@ -126,7 +132,11 @@ for name in list_vars.keys():
|
||||
src = name
|
||||
|
||||
nn = name
|
||||
nn = nn.split(".")[1:]
|
||||
if name != "proj_out.weight":
|
||||
nn = nn.split(".")[1:]
|
||||
else:
|
||||
nn = nn.split(".")
|
||||
|
||||
if nn[1] == "layers":
|
||||
nn[1] = "blocks"
|
||||
if ".".join(nn[3:-1]) == "self_attn.k_proj":
|
||||
|
Loading…
Reference in New Issue
Block a user