accept floating-point alpha value on the command line (#3712)

This commit is contained in:
Cebtenzzre 2023-08-27 17:54:43 -04:00 committed by GitHub
parent 0986868b1b
commit 2f5d769a8d
WARNING! Although there is a key with this ID in the database it does not verify this commit! This commit is SUSPICIOUS.
GPG key ID: 4AEE18F83AFDEB23

View file

@ -159,7 +159,7 @@ parser.add_argument('--rwkv-strategy', type=str, default=None, help='RWKV: The s
parser.add_argument('--rwkv-cuda-on', action='store_true', help='RWKV: Compile the CUDA kernel for better performance.') parser.add_argument('--rwkv-cuda-on', action='store_true', help='RWKV: Compile the CUDA kernel for better performance.')
# RoPE # RoPE
parser.add_argument('--alpha_value', type=int, default=1, help="Positional embeddings alpha factor for NTK RoPE scaling. Use either this or compress_pos_emb, not both.") parser.add_argument('--alpha_value', type=float, default=1, help="Positional embeddings alpha factor for NTK RoPE scaling. Use either this or compress_pos_emb, not both.")
parser.add_argument('--rope_freq_base', type=int, default=0, help="If greater than 0, will be used instead of alpha_value. Those two are related by rope_freq_base = 10000 * alpha_value ^ (64 / 63).") parser.add_argument('--rope_freq_base', type=int, default=0, help="If greater than 0, will be used instead of alpha_value. Those two are related by rope_freq_base = 10000 * alpha_value ^ (64 / 63).")
parser.add_argument('--compress_pos_emb', type=int, default=1, help="Positional embeddings compression factor. Should be set to (context length) / (model\'s original context length). Equal to 1/rope_freq_scale.") parser.add_argument('--compress_pos_emb', type=int, default=1, help="Positional embeddings compression factor. Should be set to (context length) / (model\'s original context length). Equal to 1/rope_freq_scale.")