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

cont

Browse source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2024-12-16 19:33:35 +02:00
parent 980d631032
commit 35259e5335
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
2 changed files with 127 additions and 136 deletions
Download patch file Download diff file Expand all files Collapse all files

4
examples/tts/convert_pt_to_hf.py
View file

@ -172,10 +172,6 @@ config = {
"n_embd": 768, "n_embd": 768,
"n_layer": 12 "n_layer": 12
}, },
#"n_embd_posnet": 768,
#"n_embd_convnext": 768,
#"n_layer_posnet": 6,
#"n_layer_convnext": 12
} }
with open(path_dst + '/config.json', 'w') as f: with open(path_dst + '/config.json', 'w') as f:

259
src/llama.cpp
View file

@ -2747,191 +2747,185 @@ struct llama_cparams {
struct llama_layer_posnet { struct llama_layer_posnet {
// resnet // resnet
struct ggml_tensor * norm1 = nullptr; struct ggml_tensor * norm1 = nullptr;
struct ggml_tensor * norm1_b = nullptr; struct ggml_tensor * norm1_b = nullptr;
struct ggml_tensor * conv1 = nullptr; struct ggml_tensor * conv1 = nullptr;
struct ggml_tensor * conv1_b = nullptr; struct ggml_tensor * conv1_b = nullptr;
struct ggml_tensor * norm2 = nullptr; struct ggml_tensor * norm2 = nullptr;
struct ggml_tensor * norm2_b = nullptr; struct ggml_tensor * norm2_b = nullptr;
struct ggml_tensor * conv2 = nullptr; struct ggml_tensor * conv2 = nullptr;
struct ggml_tensor * conv2_b = nullptr; struct ggml_tensor * conv2_b = nullptr;
// attention // attention
struct ggml_tensor * attn_norm = nullptr; struct ggml_tensor * attn_norm = nullptr;
struct ggml_tensor * attn_norm_b = nullptr; struct ggml_tensor * attn_norm_b = nullptr;
struct ggml_tensor * attn_q = nullptr; struct ggml_tensor * attn_q = nullptr;
struct ggml_tensor * attn_q_b = nullptr; struct ggml_tensor * attn_q_b = nullptr;
struct ggml_tensor * attn_k = nullptr; struct ggml_tensor * attn_k = nullptr;
struct ggml_tensor * attn_k_b = nullptr; struct ggml_tensor * attn_k_b = nullptr;
struct ggml_tensor * attn_v = nullptr; struct ggml_tensor * attn_v = nullptr;
struct ggml_tensor * attn_v_b = nullptr; struct ggml_tensor * attn_v_b = nullptr;
struct ggml_tensor * attn_o = nullptr; struct ggml_tensor * attn_o = nullptr;
struct ggml_tensor * attn_o_b = nullptr; struct ggml_tensor * attn_o_b = nullptr;
// normalize // normalize
struct ggml_tensor * norm = nullptr; struct ggml_tensor * norm = nullptr;
struct ggml_tensor * norm_b = nullptr; struct ggml_tensor * norm_b = nullptr;
}; };
struct llama_layer_convnext { struct llama_layer_convnext {
struct ggml_tensor * dw; struct ggml_tensor * dw = nullptr;
struct ggml_tensor * dw_b; struct ggml_tensor * dw_b = nullptr;
struct ggml_tensor * norm; struct ggml_tensor * norm = nullptr;
struct ggml_tensor * norm_b; struct ggml_tensor * norm_b = nullptr;
struct ggml_tensor * pw1; struct ggml_tensor * pw1 = nullptr;
struct ggml_tensor * pw1_b; struct ggml_tensor * pw1_b = nullptr;
struct ggml_tensor * pw2; struct ggml_tensor * pw2 = nullptr;
struct ggml_tensor * pw2_b; struct ggml_tensor * pw2_b = nullptr;
struct ggml_tensor * gamma; struct ggml_tensor * gamma = nullptr;
}; };
// TODO: separate into "llama_layer_enc" and "llama_layer_dec"
struct llama_layer { struct llama_layer {
llama_layer() {
// initialize all pointers to NULL
std::memset(this, 0, sizeof(*this));
}
// normalization // normalization
struct ggml_tensor * attn_norm; struct ggml_tensor * attn_norm = nullptr;
struct ggml_tensor * attn_norm_b; struct ggml_tensor * attn_norm_b = nullptr;
struct ggml_tensor * attn_norm_2; struct ggml_tensor * attn_norm_2 = nullptr;
struct ggml_tensor * attn_norm_2_b; struct ggml_tensor * attn_norm_2_b = nullptr;
struct ggml_tensor * attn_q_norm; struct ggml_tensor * attn_q_norm = nullptr;
struct ggml_tensor * attn_q_norm_b; struct ggml_tensor * attn_q_norm_b = nullptr;
struct ggml_tensor * attn_k_norm; struct ggml_tensor * attn_k_norm = nullptr;
struct ggml_tensor * attn_k_norm_b; struct ggml_tensor * attn_k_norm_b = nullptr;
struct ggml_tensor * attn_out_norm; struct ggml_tensor * attn_out_norm = nullptr;
struct ggml_tensor * attn_out_norm_b; struct ggml_tensor * attn_out_norm_b = nullptr;
struct ggml_tensor * attn_q_a_norm; struct ggml_tensor * attn_q_a_norm = nullptr;
struct ggml_tensor * attn_kv_a_norm; struct ggml_tensor * attn_kv_a_norm = nullptr;
struct ggml_tensor * attn_sub_norm; struct ggml_tensor * attn_sub_norm = nullptr;
struct ggml_tensor * attn_post_norm; struct ggml_tensor * attn_post_norm = nullptr;
struct ggml_tensor * ffn_sub_norm; struct ggml_tensor * ffn_sub_norm = nullptr;
struct ggml_tensor * attn_norm_cross; struct ggml_tensor * attn_norm_cross = nullptr;
struct ggml_tensor * attn_norm_enc; struct ggml_tensor * attn_norm_enc = nullptr;
// attention // attention
struct ggml_tensor * wq; struct ggml_tensor * wq = nullptr;
struct ggml_tensor * wk; struct ggml_tensor * wk = nullptr;
struct ggml_tensor * wv; struct ggml_tensor * wv = nullptr;
struct ggml_tensor * wo; struct ggml_tensor * wo = nullptr;
struct ggml_tensor * wqkv; struct ggml_tensor * wqkv = nullptr;
struct ggml_tensor * wq_a; struct ggml_tensor * wq_a = nullptr;
struct ggml_tensor * wq_b; struct ggml_tensor * wq_b = nullptr;
struct ggml_tensor * wkv_a_mqa; struct ggml_tensor * wkv_a_mqa = nullptr;
struct ggml_tensor * wkv_b; struct ggml_tensor * wkv_b = nullptr;
struct ggml_tensor * wq_cross; struct ggml_tensor * wq_cross = nullptr;
struct ggml_tensor * wk_cross; struct ggml_tensor * wk_cross = nullptr;
struct ggml_tensor * wv_cross; struct ggml_tensor * wv_cross = nullptr;
struct ggml_tensor * wo_cross; struct ggml_tensor * wo_cross = nullptr;
struct ggml_tensor * wq_enc; struct ggml_tensor * wq_enc = nullptr;
struct ggml_tensor * wk_enc; struct ggml_tensor * wk_enc = nullptr;
struct ggml_tensor * wv_enc; struct ggml_tensor * wv_enc = nullptr;
struct ggml_tensor * wo_enc; struct ggml_tensor * wo_enc = nullptr;
// attention bias // attention bias
struct ggml_tensor * bq; struct ggml_tensor * bq = nullptr;
struct ggml_tensor * bk; struct ggml_tensor * bk = nullptr;
struct ggml_tensor * bv; struct ggml_tensor * bv = nullptr;
struct ggml_tensor * bo; struct ggml_tensor * bo = nullptr;
struct ggml_tensor * bqkv; struct ggml_tensor * bqkv = nullptr;
// relative position bias // relative position bias
struct ggml_tensor * attn_rel_b; struct ggml_tensor * attn_rel_b = nullptr;
struct ggml_tensor * attn_rel_b_enc; struct ggml_tensor * attn_rel_b_enc = nullptr;
struct ggml_tensor * attn_rel_b_cross; struct ggml_tensor * attn_rel_b_cross = nullptr;
// normalization // normalization
struct ggml_tensor * ffn_norm; struct ggml_tensor * ffn_norm = nullptr;
struct ggml_tensor * ffn_norm_b; struct ggml_tensor * ffn_norm_b = nullptr;
struct ggml_tensor * ffn_post_norm; struct ggml_tensor * ffn_post_norm = nullptr;
struct ggml_tensor * layer_out_norm; struct ggml_tensor * layer_out_norm = nullptr;
struct ggml_tensor * layer_out_norm_b; struct ggml_tensor * layer_out_norm_b = nullptr;
struct ggml_tensor * ffn_norm_exps; struct ggml_tensor * ffn_norm_exps = nullptr;
struct ggml_tensor * ffn_norm_enc; struct ggml_tensor * ffn_norm_enc = nullptr;
// ff // ff
struct ggml_tensor * ffn_gate; // w1 struct ggml_tensor * ffn_gate = nullptr; // w1
struct ggml_tensor * ffn_down; // w2 struct ggml_tensor * ffn_down = nullptr; // w2
struct ggml_tensor * ffn_up; // w3 struct ggml_tensor * ffn_up = nullptr; // w3
struct ggml_tensor * ffn_gate_enc; struct ggml_tensor * ffn_gate_enc = nullptr;
struct ggml_tensor * ffn_down_enc; struct ggml_tensor * ffn_down_enc = nullptr;
struct ggml_tensor * ffn_up_enc; struct ggml_tensor * ffn_up_enc = nullptr;
// ff MoE // ff MoE
struct ggml_tensor * ffn_gate_inp; struct ggml_tensor * ffn_gate_inp = nullptr;
struct ggml_tensor * ffn_gate_exps; struct ggml_tensor * ffn_gate_exps = nullptr;
struct ggml_tensor * ffn_down_exps; struct ggml_tensor * ffn_down_exps = nullptr;
struct ggml_tensor * ffn_up_exps ; struct ggml_tensor * ffn_up_exps = nullptr;
// ff shared expert (shexp) // ff shared expert (shexp)
struct ggml_tensor * ffn_gate_inp_shexp; struct ggml_tensor * ffn_gate_inp_shexp = nullptr;
struct ggml_tensor * ffn_gate_shexp; struct ggml_tensor * ffn_gate_shexp = nullptr;
struct ggml_tensor * ffn_down_shexp; struct ggml_tensor * ffn_down_shexp = nullptr;
struct ggml_tensor * ffn_up_shexp; struct ggml_tensor * ffn_up_shexp = nullptr;
// ff bias // ff bias
struct ggml_tensor * ffn_gate_b; struct ggml_tensor * ffn_gate_b = nullptr;
struct ggml_tensor * ffn_down_b; // b2 struct ggml_tensor * ffn_down_b = nullptr; // b2
struct ggml_tensor * ffn_up_b; // b3 struct ggml_tensor * ffn_up_b = nullptr; // b3
struct ggml_tensor * ffn_act; struct ggml_tensor * ffn_act = nullptr;
// mamba proj // mamba proj
struct ggml_tensor * ssm_in; struct ggml_tensor * ssm_in = nullptr;
struct ggml_tensor * ssm_x; struct ggml_tensor * ssm_x = nullptr;
struct ggml_tensor * ssm_dt; struct ggml_tensor * ssm_dt = nullptr;
struct ggml_tensor * ssm_out; struct ggml_tensor * ssm_out = nullptr;
// mamba // mamba
struct ggml_tensor * ssm_conv1d; struct ggml_tensor * ssm_conv1d = nullptr;
struct ggml_tensor * ssm_a; struct ggml_tensor * ssm_a = nullptr;
struct ggml_tensor * ssm_d; struct ggml_tensor * ssm_d = nullptr;
// mamba bias // mamba bias
struct ggml_tensor * ssm_conv1d_b; struct ggml_tensor * ssm_conv1d_b = nullptr;
struct ggml_tensor * ssm_dt_b; struct ggml_tensor * ssm_dt_b = nullptr;
// rwkv // rwkv
struct ggml_tensor * time_mix_w1; struct ggml_tensor * time_mix_w1 = nullptr;
struct ggml_tensor * time_mix_w2; struct ggml_tensor * time_mix_w2 = nullptr;
struct ggml_tensor * time_mix_lerp_x; struct ggml_tensor * time_mix_lerp_x = nullptr;
struct ggml_tensor * time_mix_lerp_w; struct ggml_tensor * time_mix_lerp_w = nullptr;
struct ggml_tensor * time_mix_lerp_k; struct ggml_tensor * time_mix_lerp_k = nullptr;
struct ggml_tensor * time_mix_lerp_v; struct ggml_tensor * time_mix_lerp_v = nullptr;
struct ggml_tensor * time_mix_lerp_r; struct ggml_tensor * time_mix_lerp_r = nullptr;
struct ggml_tensor * time_mix_lerp_g; struct ggml_tensor * time_mix_lerp_g = nullptr;
struct ggml_tensor * time_mix_first; struct ggml_tensor * time_mix_first = nullptr;
struct ggml_tensor * time_mix_decay; struct ggml_tensor * time_mix_decay = nullptr;
struct ggml_tensor * time_mix_decay_w1; struct ggml_tensor * time_mix_decay_w1 = nullptr;
struct ggml_tensor * time_mix_decay_w2; struct ggml_tensor * time_mix_decay_w2 = nullptr;
struct ggml_tensor * time_mix_key; struct ggml_tensor * time_mix_key = nullptr;
struct ggml_tensor * time_mix_value; struct ggml_tensor * time_mix_value = nullptr;
struct ggml_tensor * time_mix_receptance; struct ggml_tensor * time_mix_receptance = nullptr;
struct ggml_tensor * time_mix_gate; struct ggml_tensor * time_mix_gate = nullptr;
struct ggml_tensor * time_mix_ln; struct ggml_tensor * time_mix_ln = nullptr;
struct ggml_tensor * time_mix_ln_b; struct ggml_tensor * time_mix_ln_b = nullptr;
struct ggml_tensor * time_mix_output; struct ggml_tensor * time_mix_output = nullptr;
struct ggml_tensor * channel_mix_lerp_k; struct ggml_tensor * channel_mix_lerp_k = nullptr;
struct ggml_tensor * channel_mix_lerp_r; struct ggml_tensor * channel_mix_lerp_r = nullptr;
struct ggml_tensor * channel_mix_key; struct ggml_tensor * channel_mix_key = nullptr;
struct ggml_tensor * channel_mix_receptance; struct ggml_tensor * channel_mix_receptance = nullptr;
struct ggml_tensor * channel_mix_value; struct ggml_tensor * channel_mix_value = nullptr;
// long rope factors // long rope factors
struct ggml_tensor * rope_long = nullptr; struct ggml_tensor * rope_long = nullptr;
@ -2939,13 +2933,13 @@ struct llama_layer {
struct ggml_tensor * rope_freqs = nullptr; struct ggml_tensor * rope_freqs = nullptr;
// bitnet scale // bitnet scale
struct ggml_tensor * wq_scale; struct ggml_tensor * wq_scale = nullptr;
struct ggml_tensor * wk_scale; struct ggml_tensor * wk_scale = nullptr;
struct ggml_tensor * wv_scale; struct ggml_tensor * wv_scale = nullptr;
struct ggml_tensor * wo_scale; struct ggml_tensor * wo_scale = nullptr;
struct ggml_tensor * ffn_gate_scale; struct ggml_tensor * ffn_gate_scale = nullptr;
struct ggml_tensor * ffn_up_scale; struct ggml_tensor * ffn_up_scale = nullptr;
struct ggml_tensor * ffn_down_scale; struct ggml_tensor * ffn_down_scale = nullptr;
struct llama_layer_posnet posnet; struct llama_layer_posnet posnet;
@ -3167,6 +3161,7 @@ struct llama_sbatch {
// batch indices of the output // batch indices of the output
std::vector<size_t> out_ids; std::vector<size_t> out_ids;
std::vector<llama_sbatch_seq> seq; std::vector<llama_sbatch_seq> seq;
const llama_batch * batch = nullptr; const llama_batch * batch = nullptr;
// buffers for the ubatch // buffers for the ubatch