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

llama : implement YaRN RoPE scaling (#2268)

Browse source 
Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
Co-authored-by: Jeffrey Quesnelle <jquesnelle@gmail.com>
This commit is contained in:
cebtenzzre 2023-11-01 18:04:33 -04:00 • committed by GitHub
parent c43c2da8af
commit 898aeca90a
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
15 changed files with 763 additions and 257 deletions
Download patch file Download diff file Expand all files Collapse all files

6
examples/train-text-from-scratch/train-text-from-scratch.cpp
View file

@ -349,9 +349,9 @@ static struct ggml_tensor * llama_build_train_graphs(
// not capturing these, to silcence warnings
const int rope_mode = 0;
return ggml_rope_custom(ctx,
t, KQ_pos, n_rot, rope_mode, n_ctx,
rope_freq_base, rope_freq_scale);
return ggml_rope_custom(
ctx, t, KQ_pos, n_rot, rope_mode, n_ctx, 0, rope_freq_base, rope_freq_scale, 0.0f, 1.0f, 0.0f, 0.0f
);
};
set_name(tokens_input, "tokens_input");