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

Fixed all models' control vectors

Browse source
This commit is contained in:
jukofyork 2024-06-22 17:10:08 +01:00 • committed by GitHub
parent 2c42c51ad5
commit f57ebf281e
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 111 additions and 32 deletions
Download patch file Download diff file Expand all files Collapse all files

143
llama.cpp
View file

@ -7818,8 +7818,7 @@ struct llm_build_context {
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
cb(cur, "ffn_out", il); cb(cur, "ffn_out", il);
ggml_tensor * layer_dir = lctx.cvec.tensor_for(il); if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
if (layer_dir != nullptr) {
cur = ggml_add(ctx0, cur, layer_dir); cur = ggml_add(ctx0, cur, layer_dir);
} }
cb(cur, "l_out", il); cb(cur, "l_out", il);
@ -7936,6 +7935,9 @@ struct llm_build_context {
} }
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -8040,6 +8042,9 @@ struct llm_build_context {
} }
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -8158,6 +8163,9 @@ struct llm_build_context {
cb(cur, "l_out", il); cb(cur, "l_out", il);
cur = ggml_add(ctx0, cur, inpL); cur = ggml_add(ctx0, cur, inpL);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -8309,8 +8317,7 @@ struct llm_build_context {
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
cb(cur, "ffn_out", il); cb(cur, "ffn_out", il);
ggml_tensor * layer_dir = lctx.cvec.tensor_for(il); if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
if (layer_dir != nullptr) {
cur = ggml_add(ctx0, cur, layer_dir); cur = ggml_add(ctx0, cur, layer_dir);
} }
cb(cur, "l_out", il); cb(cur, "l_out", il);
@ -8443,8 +8450,7 @@ struct llm_build_context {
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
cb(cur, "ffn_out", il); cb(cur, "ffn_out", il);
ggml_tensor * layer_dir = lctx.cvec.tensor_for(il); if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
if (layer_dir != nullptr) {
cur = ggml_add(ctx0, cur, layer_dir); cur = ggml_add(ctx0, cur, layer_dir);
} }
cb(cur, "l_out", il); cb(cur, "l_out", il);
@ -8552,8 +8558,14 @@ struct llm_build_context {
cb(cur, "ffn_out", il); cb(cur, "ffn_out", il);
} }
inpL = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
cb(inpL, "l_out", il); if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il);
// input for next layer
inpL = cur;
} }
cur = llm_build_norm(ctx0, inpL, hparams, cur = llm_build_norm(ctx0, inpL, hparams,
@ -8641,6 +8653,9 @@ struct llm_build_context {
} }
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -8936,8 +8951,14 @@ struct llm_build_context {
cb(cur, "ffn_out", il); cb(cur, "ffn_out", il);
} }
inpL = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
cb(inpL, "l_out", il); if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il);
// input for next layer
inpL = cur;
} }
cur = llm_build_norm(ctx0, inpL, hparams, cur = llm_build_norm(ctx0, inpL, hparams,
@ -9071,6 +9092,9 @@ struct llm_build_context {
} }
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -9219,6 +9243,9 @@ struct llm_build_context {
} }
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -9331,6 +9358,9 @@ struct llm_build_context {
} }
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -9442,13 +9472,9 @@ struct llm_build_context {
cb(cur, "ffn_out", il); cb(cur, "ffn_out", il);
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
// add control vector
ggml_tensor * layer_dir = lctx.cvec.tensor_for(il);
if (layer_dir != nullptr) {
cur = ggml_add(ctx0, cur, layer_dir); cur = ggml_add(ctx0, cur, layer_dir);
} }
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -9594,6 +9620,9 @@ struct llm_build_context {
} }
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -9717,8 +9746,12 @@ struct llm_build_context {
cb(cur, "l_out", il); cb(cur, "l_out", il);
cur = ggml_add(ctx0, cur, inpL); cur = ggml_add(ctx0, cur, inpL);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer
inpL = cur; inpL = cur;
} }
@ -9850,8 +9883,12 @@ struct llm_build_context {
} }
cur = ggml_add(ctx0, residual, cur); cur = ggml_add(ctx0, residual, cur);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer
inpL = cur; inpL = cur;
} }
@ -9953,6 +9990,9 @@ struct llm_build_context {
cb(cur, "l_out", il); cb(cur, "l_out", il);
cur = ggml_add(ctx0, cur, inpL); cur = ggml_add(ctx0, cur, inpL);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -10058,8 +10098,14 @@ struct llm_build_context {
cb(cur, "ffn_out", il); cb(cur, "ffn_out", il);
} }
inpL = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
cb(inpL, "l_out", il); if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il);
// input for next layer
inpL = cur;
} }
cur = llm_build_norm(ctx0, inpL, hparams, cur = llm_build_norm(ctx0, inpL, hparams,
@ -10165,8 +10211,14 @@ struct llm_build_context {
cb(cur, "ffn_out", il); cb(cur, "ffn_out", il);
} }
inpL = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
cb(inpL, "l_out", il); if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il);
// input for next layer
inpL = cur;
} }
cur = llm_build_norm(ctx0, inpL, hparams, cur = llm_build_norm(ctx0, inpL, hparams,
@ -10278,6 +10330,9 @@ struct llm_build_context {
cb(cur, "ffn_out", il); cb(cur, "ffn_out", il);
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -10395,6 +10450,9 @@ struct llm_build_context {
cb(cur, "ffn_out", il); cb(cur, "ffn_out", il);
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -10536,6 +10594,9 @@ struct llm_build_context {
cb(cur, "hidden_scaled_ffn", -1); cb(cur, "hidden_scaled_ffn", -1);
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -10648,6 +10709,9 @@ struct llm_build_context {
} }
cur = ggml_add(ctx0, cur, sa_out); cur = ggml_add(ctx0, cur, sa_out);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -10765,6 +10829,9 @@ struct llm_build_context {
LLM_FFN_GELU, LLM_FFN_SEQ, cb, il); LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
cb(cur, "ffn_out", il); cb(cur, "ffn_out", il);
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -10913,6 +10980,9 @@ struct llm_build_context {
// residual // residual
cur = ggml_add(ctx0, cur, inpL); cur = ggml_add(ctx0, cur, inpL);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -11054,13 +11124,9 @@ struct llm_build_context {
// add together residual + FFN + self-attention // add together residual + FFN + self-attention
cur = ggml_add(ctx0, cur, inpL); cur = ggml_add(ctx0, cur, inpL);
cur = ggml_add(ctx0, cur, attn_out); cur = ggml_add(ctx0, cur, attn_out);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
// add control vector
ggml_tensor * layer_dir = lctx.cvec.tensor_for(il);
if (layer_dir != nullptr) {
cur = ggml_add(ctx0, cur, layer_dir); cur = ggml_add(ctx0, cur, layer_dir);
} }
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer
@ -11196,8 +11262,7 @@ struct llm_build_context {
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
cb(cur, "ffn_out", il); cb(cur, "ffn_out", il);
ggml_tensor * layer_dir = lctx.cvec.tensor_for(il); if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
if (layer_dir != nullptr) {
cur = ggml_add(ctx0, cur, layer_dir); cur = ggml_add(ctx0, cur, layer_dir);
} }
cb(cur, "l_out", il); cb(cur, "l_out", il);
@ -11313,8 +11378,14 @@ struct llm_build_context {
cur = ggml_add(ctx0, cur, inpL); cur = ggml_add(ctx0, cur, inpL);
cb(cur, "ffn_out", il); cb(cur, "ffn_out", il);
inpL = ggml_add(ctx0, cur, attn_out); cur = ggml_add(ctx0, cur, attn_out);
cb(inpL, "l_out", il); if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il);
// input for next layer
inpL = cur;
} else { } else {
// attention and ffn are computed sequentially // attention and ffn are computed sequentially
// x = x + attn(ln1(x)) // x = x + attn(ln1(x))
@ -11337,8 +11408,14 @@ struct llm_build_context {
LLM_FFN_GELU, LLM_FFN_SEQ, cb, il); LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
cb(cur, "ffn_out", il); cb(cur, "ffn_out", il);
inpL = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
cb(inpL, "l_out", il); if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il);
// input for next layer
inpL = cur;
} }
} }
@ -11465,8 +11542,7 @@ struct llm_build_context {
cur = ggml_add(ctx0, cur, ffn_out); cur = ggml_add(ctx0, cur, ffn_out);
cb(cur, "ffn_out", il); cb(cur, "ffn_out", il);
ggml_tensor * layer_dir = lctx.cvec.tensor_for(il); if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
if (layer_dir != nullptr) {
cur = ggml_add(ctx0, cur, layer_dir); cur = ggml_add(ctx0, cur, layer_dir);
} }
cb(cur, "l_out", il); cb(cur, "l_out", il);
@ -11701,6 +11777,9 @@ struct llm_build_context {
} }
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
if (ggml_tensor * layer_dir = lctx.cvec.tensor_for(il)) {
cur = ggml_add(ctx0, cur, layer_dir);
}
cb(cur, "l_out", il); cb(cur, "l_out", il);
// input for next layer // input for next layer