diff --git a/modules/exllamav2.py b/modules/exllamav2.py index c64714e8..6d4603c5 100644 --- a/modules/exllamav2.py +++ b/modules/exllamav2.py @@ -28,9 +28,9 @@ class Exllamav2Model: config.prepare() config.max_seq_len = shared.args.max_seq_len - config.rope_scale = shared.args.compress_pos_emb - config.rope_alpha = shared.args.alpha_value - + config.scale_pos_emb = shared.args.compress_pos_emb + config.scale_alpha_value = shared.args.alpha_value + model = ExLlamaV2(config) split = None diff --git a/modules/exllamav2_hf.py b/modules/exllamav2_hf.py index 65b98596..a8837af1 100644 --- a/modules/exllamav2_hf.py +++ b/modules/exllamav2_hf.py @@ -116,7 +116,7 @@ class Exllamav2HF(PreTrainedModel): config.prepare() config.max_seq_len = shared.args.max_seq_len - config.rope_scale = shared.args.compress_pos_emb - config.rope_alpha = shared.args.alpha_value + config.scale_pos_emb = shared.args.compress_pos_emb + config.scale_alpha_value = shared.args.alpha_value return Exllamav2HF(config)