diff --git a/ggml/include/ggml.h b/ggml/include/ggml.h index 76a3176a1..39aff9e39 100644 --- a/ggml/include/ggml.h +++ b/ggml/include/ggml.h @@ -513,7 +513,6 @@ extern "C" { GGML_OP_GET_REL_POS, GGML_OP_ADD_REL_POS, GGML_OP_RWKV_WKV, - GGML_OP_RWKV_TOKEN_SHIFT, GGML_OP_UNARY, @@ -1905,14 +1904,7 @@ extern "C" { struct ggml_tensor * r, struct ggml_tensor * tf, struct ggml_tensor * td, - struct ggml_tensor * state, - struct ggml_tensor * state_seq); - - GGML_API struct ggml_tensor * ggml_rwkv_token_shift( - struct ggml_context * ctx, - struct ggml_tensor * x_carry, - struct ggml_tensor * x_norm, - struct ggml_tensor * state_seq); + struct ggml_tensor * state); // custom operators diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c index a32cfcb09..93f3933e7 100644 --- a/ggml/src/ggml.c +++ b/ggml/src/ggml.c @@ -2836,7 +2836,6 @@ static const char * GGML_OP_NAME[GGML_OP_COUNT] = { "GET_REL_POS", "ADD_REL_POS", "RWKV_WKV", - "RWKV_TOKEN_SHIFT", "UNARY", @@ -2855,7 +2854,7 @@ static const char * GGML_OP_NAME[GGML_OP_COUNT] = { "CROSS_ENTROPY_LOSS_BACK", }; -static_assert(GGML_OP_COUNT == 80, "GGML_OP_COUNT != 80"); +static_assert(GGML_OP_COUNT == 79, "GGML_OP_COUNT != 79"); static const char * GGML_OP_SYMBOL[GGML_OP_COUNT] = { "none", @@ -2929,8 +2928,7 @@ static const char * GGML_OP_SYMBOL[GGML_OP_COUNT] = { "win_unpart(x)", "get_rel_pos(x)", "add_rel_pos(x)", - "rwkv_wkv(k, v, r, tf, td, s, sq)", - "rwkv_token_shift(xc, xn, sq)", + "rwkv_wkv(k, v, r, tf, td, s)", "unary(x)", @@ -2949,7 +2947,7 @@ static const char * GGML_OP_SYMBOL[GGML_OP_COUNT] = { "cross_entropy_loss_back(x,y)", }; -static_assert(GGML_OP_COUNT == 80, "GGML_OP_COUNT != 80"); +static_assert(GGML_OP_COUNT == 79, "GGML_OP_COUNT != 79"); static_assert(GGML_OP_POOL_COUNT == 2, "GGML_OP_POOL_COUNT != 2"); @@ -7650,39 +7648,36 @@ struct ggml_tensor * ggml_rwkv_wkv( struct ggml_tensor * r, struct ggml_tensor * tf, struct ggml_tensor * td, - struct ggml_tensor * state, - struct ggml_tensor * state_seq) { + struct ggml_tensor * state) { GGML_ASSERT(ggml_is_contiguous(k)); GGML_ASSERT(ggml_is_contiguous(v)); GGML_ASSERT(ggml_is_contiguous(r)); GGML_ASSERT(ggml_is_contiguous(tf)); GGML_ASSERT(ggml_is_contiguous(td)); GGML_ASSERT(ggml_is_contiguous(state)); - GGML_ASSERT(ggml_is_contiguous(state_seq)); - GGML_ASSERT(state_seq->type == GGML_TYPE_I32); const int64_t S = k->ne[0]; const int64_t H = k->ne[2]; const int64_t n_tokens = k->ne[3]; - const int64_t n_kv = state_seq->ne[0]; + const int64_t n_seqs = state->ne[1]; { GGML_ASSERT(k->ne[1] == 1); GGML_ASSERT(v->ne[0] == 1 && v->ne[1] == S && v->ne[2] == H && v->ne[3] == n_tokens); GGML_ASSERT(r->ne[0] == 1 && r->ne[1] == S && r->ne[2] == H && r->ne[3] == n_tokens); // TODO: RWKV v4 and v5 GGML_ASSERT(td->ne[0] == 1 && td->ne[1] == S && td->ne[2] == H && td->ne[3] == n_tokens); - GGML_ASSERT(ggml_nelements(state) == S * S * H * n_kv); + GGML_ASSERT(ggml_nelements(state) == S * S * H * n_seqs); } bool is_node = false; - if (k->grad || v->grad || r->grad || tf->grad || td->grad || state->grad || state_seq->grad) { + if (k->grad || v->grad || r->grad || tf->grad || td->grad || state->grad) { GGML_ABORT("fatal error"); // TODO: implement backward is_node = true; } // concat output and new_state - const int64_t ne[4] = { S * H, n_tokens + S * n_kv, 1, 1 }; + const int64_t ne[4] = { S * H, n_tokens + S * n_seqs, 1, 1 }; struct ggml_tensor * result = ggml_new_tensor(ctx, GGML_TYPE_F32, 4, ne); result->op = GGML_OP_RWKV_WKV; @@ -7693,48 +7688,6 @@ struct ggml_tensor * ggml_rwkv_wkv( result->src[3] = tf; result->src[4] = td; result->src[5] = state; - result->src[6] = state_seq; - - return result; -} - -// ggml_rwkv_token_shift - -struct ggml_tensor * ggml_rwkv_token_shift( - struct ggml_context * ctx, - struct ggml_tensor * x_carry, - struct ggml_tensor * x_norm, - struct ggml_tensor * state_seq) { - GGML_ASSERT(ggml_is_contiguous(x_carry)); - GGML_ASSERT(ggml_is_contiguous(x_norm)); - GGML_ASSERT(ggml_is_contiguous(state_seq)); - GGML_ASSERT(state_seq->type == GGML_TYPE_I32); - - const int64_t n_embd = x_norm->ne[0]; - const int64_t n_kv = state_seq->ne[0]; - const int64_t n_tokens = state_seq->ne[1]; - { - GGML_ASSERT(x_norm->ne[0] == n_embd); - GGML_ASSERT(x_norm->ne[1] == n_tokens); - GGML_ASSERT(ggml_nelements(x_carry) == n_embd * n_kv); - } - - bool is_node = false; - - if (x_carry->grad || x_norm->grad || state_seq->grad) { - GGML_ABORT("fatal error"); // TODO: implement backward - is_node = true; - } - - // concat output and new_state - const int64_t ne[4] = { n_embd, n_tokens + n_kv, 1, 1 }; - struct ggml_tensor * result = ggml_new_tensor(ctx, GGML_TYPE_F32, 4, ne); - - result->op = GGML_OP_RWKV_TOKEN_SHIFT; - result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL; - result->src[0] = x_carry; - result->src[1] = x_norm; - result->src[2] = state_seq; return result; } @@ -16905,7 +16858,7 @@ static void ggml_compute_forward_rwkv_wkv_f32( const size_t T = dst->src[1]->ne[3]; const size_t C = dst->ne[0]; const size_t H = dst->src[1]->ne[2]; - const size_t n_kv = dst->src[6]->ne[0]; + const size_t n_seqs = dst->src[5]->ne[1]; float * dst_data = (float *) dst->data; float * state = ((float *) dst->data) + C * T; @@ -16921,8 +16874,7 @@ static void ggml_compute_forward_rwkv_wkv_f32( float * r = (float *) dst->src[2]->data; float * time_faaaa = (float *) dst->src[3]->data; float * time_decay = (float *) dst->src[4]->data; - int32_t * seq_data = (int32_t *) dst->src[6]->data; - memcpy(state, dst->src[5]->data, (C / H) * C * n_kv * sizeof(float)); + memcpy(state, dst->src[5]->data, (C / H) * C * n_seqs * sizeof(float)); size_t t_stride = H * (C / H); @@ -16935,7 +16887,7 @@ static void ggml_compute_forward_rwkv_wkv_f32( // recursive through each token for (size_t t = 0; t < T; t++) { size_t t_offset = t * t_stride; - float * state_cur = state + (C / H) * C * seq_data[t * n_kv]; + float * state_cur = state + (C / H) * C * (t / (T / n_seqs)); for (size_t h = 0; h < H; h++) { size_t h_offset = h * h_stride; @@ -16967,15 +16919,6 @@ static void ggml_compute_forward_rwkv_wkv_f32( } } } - - for (size_t t = 0; t < T; t++) { - for (size_t kv = 1; kv < n_kv; kv++) { - int64_t seq = seq_data[t * n_kv + kv]; - if (seq >= 0 && seq_data[(t + 1) * n_kv + kv] != seq) { - memcpy(state + (C / H) * C * seq, state + (C / H) * C * seq_data[t * n_kv], (C / H) * C * sizeof(float)); - } - } - } } static void ggml_compute_forward_rwkv_wkv( @@ -16996,77 +16939,6 @@ static void ggml_compute_forward_rwkv_wkv( } } -static void ggml_compute_forward_rwkv_token_shift_f32( - const struct ggml_compute_params * params, - struct ggml_tensor * dst) { - const int64_t n_embd = dst->ne[0]; - const int64_t n_kv = dst->src[2]->ne[0]; - const int64_t n_tokens = dst->src[1]->ne[1]; - float * dst_data = (float *) dst->data; - float * x_carry = (float *) dst->src[0]->data; - float * x_norm = (float *) dst->src[1]->data; - int32_t * sq_data = (int32_t *) dst->src[2]->data; - - if (params->ith != 0) { - return; - } - - int32_t seq_start = 0; - int32_t seq_length = 0; - - for (int i1 = 0; i1 < n_kv; ++i1) { - seq_start = -1; - // assume that the tokens for each sequence are contiguous - for (int i2 = 0; i2 < n_tokens; ++i2) { - int32_t seq = sq_data[i2*n_kv]; - if (seq == i1 && seq_start < 0) { - seq_start = i2; - } - - if ((seq_start >= 0 && seq != i1) || i2 == n_tokens - 1) { - seq_length = i2 - seq_start + (i2 == n_tokens - 1); - break; - } - } - - if (seq_start >= 0) { - int32_t seq = sq_data[seq_start*n_kv]; - memcpy(dst_data + seq_start*n_embd, x_carry + seq*n_embd, n_embd*sizeof(float)); - memcpy(dst_data + (seq_start+1)*n_embd, x_norm + seq_start*n_embd, (seq_length-1)*n_embd*sizeof(float)); - } - } - - for (int i3 = 0; i3 < n_kv; ++i3) { - int32_t last_token_pos = 0; - for (int i4 = 0; i4 < n_tokens; ++i4) { - for (int i5 = 0; i5 < n_kv; ++i5) { - if (sq_data[i4*n_kv + i5] == i3) { - last_token_pos = i4; - } - } - } - memcpy(dst_data + (n_tokens + i3)*n_embd, x_norm + last_token_pos*n_embd, n_embd*sizeof(float)); - } -} - -static void ggml_compute_forward_rwkv_token_shift( - const struct ggml_compute_params * params, - struct ggml_tensor * dst) { - - const struct ggml_tensor * src0 = dst->src[0]; - - switch (src0->type) { - case GGML_TYPE_F32: - { - ggml_compute_forward_rwkv_token_shift_f32(params, dst); - } break; - default: - { - GGML_ABORT("fatal error"); - } - } -} - // ggml_compute_forward_map_unary static void ggml_compute_forward_map_unary_f32( @@ -17722,10 +17594,6 @@ static void ggml_compute_forward(struct ggml_compute_params * params, struct ggm { ggml_compute_forward_rwkv_wkv(params, tensor); } break; - case GGML_OP_RWKV_TOKEN_SHIFT: - { - ggml_compute_forward_rwkv_token_shift(params, tensor); - } break; case GGML_OP_MAP_UNARY: { ggml_unary_op_f32_t fun; @@ -18859,7 +18727,6 @@ static void ggml_compute_backward(struct ggml_context * ctx, struct ggml_tensor case GGML_OP_GET_REL_POS: case GGML_OP_ADD_REL_POS: case GGML_OP_RWKV_WKV: - case GGML_OP_RWKV_TOKEN_SHIFT: case GGML_OP_MAP_UNARY: case GGML_OP_MAP_BINARY: case GGML_OP_MAP_CUSTOM1_F32: @@ -19435,7 +19302,6 @@ static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads) { case GGML_OP_WIN_UNPART: case GGML_OP_GET_REL_POS: case GGML_OP_RWKV_WKV: - case GGML_OP_RWKV_TOKEN_SHIFT: case GGML_OP_MAP_UNARY: case GGML_OP_MAP_BINARY: case GGML_OP_MAP_CUSTOM1_F32: diff --git a/src/llama.cpp b/src/llama.cpp index a6f6ef124..f8ec0e323 100644 --- a/src/llama.cpp +++ b/src/llama.cpp @@ -9378,15 +9378,20 @@ static struct ggml_tensor * llm_build_time_mix_rwkv6( const struct llama_layer * layer, struct ggml_tensor * cur, struct ggml_tensor * x_prev, - struct ggml_tensor ** wkv_state, - struct ggml_tensor * state_seq) { + struct ggml_tensor ** wkv_state) { size_t n_embed = cur->ne[0]; - size_t n_tokens = cur->ne[1]; + size_t n_seq_tokens = cur->ne[1]; + size_t n_seqs = cur->ne[2]; size_t head_size = layer->time_mix_first->ne[0]; size_t head_count = layer->time_mix_first->ne[1]; - size_t n_kv = state_seq->ne[0]; + + size_t n_tokens = n_seqs * n_seq_tokens; struct ggml_tensor * sx = ggml_sub(ctx, x_prev, cur); + + sx = ggml_reshape_2d(ctx, sx, n_embed, n_tokens); + cur = ggml_reshape_2d(ctx, cur, n_embed, n_tokens); + struct ggml_tensor * xxx = ggml_add(ctx, ggml_mul(ctx, sx, layer->time_mix_lerp_x), cur); xxx = ggml_reshape_4d( @@ -9489,9 +9494,9 @@ static struct ggml_tensor * llm_build_time_mix_rwkv6( k = ggml_transpose(ctx, k); v = ggml_transpose(ctx, v); r = ggml_transpose(ctx, r); - struct ggml_tensor * wkv_output = ggml_rwkv_wkv(ctx, k, v, r, layer->time_mix_first, w, *wkv_state, state_seq); + struct ggml_tensor * wkv_output = ggml_rwkv_wkv(ctx, k, v, r, layer->time_mix_first, w, *wkv_state); cur = ggml_view_1d(ctx, wkv_output, n_embed * n_tokens, 0); - *wkv_state = ggml_view_1d(ctx, wkv_output, n_embed * head_size * n_kv, n_embed * n_tokens * sizeof(float)); + *wkv_state = ggml_view_1d(ctx, wkv_output, n_embed * head_size * n_seqs, n_embed * n_tokens * sizeof(float)); // ggml_group_norm considers groups in the third dimension. cur = ggml_reshape_4d(ctx, cur, n_embed / head_count, 1, head_count, n_tokens); @@ -9501,8 +9506,9 @@ static struct ggml_tensor * llm_build_time_mix_rwkv6( cur = ggml_add(ctx, ggml_mul(ctx, cur, layer->time_mix_ln), layer->time_mix_ln_b); cur = ggml_mul(ctx, cur, g); + cur = ggml_mul_mat(ctx, layer->time_mix_output, cur); - return ggml_mul_mat(ctx, layer->time_mix_output, cur); + return ggml_reshape_3d(ctx, cur, n_embed, n_seq_tokens, n_seqs); } static struct ggml_tensor * llm_build_channel_mix_rwkv6( @@ -15053,49 +15059,56 @@ struct llm_build_context { // Token shift state dimensions should be 2 * n_emb GGML_ASSERT(n_embd == hparams.n_embd_k_s() / 2); + const int64_t n_seqs = batch.n_seqs; + const int64_t n_seq_tokens = batch.n_seq_tokens; + const int64_t n_tokens = batch.n_tokens; + GGML_ASSERT(n_seqs != 0); + GGML_ASSERT(batch.equal_seqs); + GGML_ASSERT(n_tokens == n_seq_tokens * n_seqs); + ggml_tensor * input_embeddings = llm_build_inp_embd(ctx0, lctx, hparams, batch, model.tok_embd, cb); + struct ggml_tensor * state_copy = build_inp_s_copy(); struct ggml_tensor * state_mask = build_inp_s_mask(); - struct ggml_tensor * state_seq = build_inp_s_seq(); ggml_tensor * cur = llm_build_norm(ctx0, input_embeddings, hparams, model.tok_norm, model.tok_norm_b, LLM_NORM, cb, -1); for (int layer_i = 0; layer_i < n_layer; ++layer_i) { const llama_layer * layer = &model.layers[layer_i]; - struct ggml_tensor * token_shift = ggml_reshape_2d(ctx0, kv_self.k_l[layer_i], hparams.n_embd_k_s(), kv_self.size); - struct ggml_tensor * wkv_states = ggml_reshape_2d(ctx0, kv_self.v_l[layer_i], hparams.n_embd_v_s(), kv_self.size); + // (ab)using the KV cache to store the states + struct ggml_tensor * token_shift = llm_build_copy_mask_state(ctx0, + gf, kv_self.k_l[layer_i], state_copy, state_mask, + hparams.n_embd_k_s(), kv_self.size, kv_head, n_kv, n_seqs); + struct ggml_tensor * wkv_states = llm_build_copy_mask_state(ctx0, + gf, kv_self.v_l[layer_i], state_copy, state_mask, + hparams.n_embd_v_s(), kv_self.size, kv_head, n_kv, n_seqs); - { - token_shift = ggml_mul(ctx0, - ggml_view_2d(ctx0, token_shift, token_shift->ne[0], n_kv, token_shift->nb[1], kv_head*token_shift->nb[1]), - state_mask); - wkv_states = ggml_mul(ctx0, - ggml_view_2d(ctx0, wkv_states, wkv_states->ne[0], n_kv, wkv_states->nb[1], kv_head*wkv_states->nb[1]), - state_mask); - } + cur = ggml_reshape_3d(ctx0, cur, n_embd, n_seq_tokens, n_seqs); token_shift = ggml_cont( ctx0, ggml_permute( ctx0, - ggml_reshape_3d(ctx0, token_shift, n_embd, 2, n_kv), + ggml_reshape_3d(ctx0, token_shift, n_embd, 2, n_seqs), 0, 2, 1, 3 ) ); - struct ggml_tensor * att_shift = ggml_view_1d(ctx0, token_shift, n_embd * n_kv, 0); - struct ggml_tensor * ffn_shift = ggml_view_1d(ctx0, token_shift, n_embd * n_kv, n_embd * n_kv * ggml_element_size(kv_self.k_l[layer_i])); + struct ggml_tensor * att_shift = ggml_view_1d(ctx0, token_shift, n_embd * n_seqs, 0); + struct ggml_tensor * ffn_shift = ggml_view_1d(ctx0, token_shift, n_embd * n_seqs, n_embd * n_seqs * ggml_element_size(token_shift)); + att_shift = ggml_reshape_3d(ctx0, att_shift, n_embd, 1, n_seqs); + ffn_shift = ggml_reshape_3d(ctx0, ffn_shift, n_embd, 1, n_seqs); struct ggml_tensor * x_norm = llm_build_norm(ctx0, cur, hparams, layer->attn_norm, layer->attn_norm_b, LLM_NORM, cb, layer_i); - struct ggml_tensor * tmp = ggml_rwkv_token_shift(ctx0, att_shift, x_norm, state_seq); - struct ggml_tensor * x_prev = ggml_reshape_2d( + struct ggml_tensor * x_prev = ggml_concat( ctx0, - ggml_view_1d(ctx0, tmp, n_embd * n_tokens, 0), - n_embd, n_tokens + att_shift, + ggml_view_3d(ctx0, x_norm, n_embd, n_seq_tokens - 1, n_seqs, x_norm->nb[1], x_norm->nb[2], 0), + 1 ); - cur = ggml_add(ctx0, cur, llm_build_time_mix_rwkv6(ctx0, layer, x_norm, x_prev, &wkv_states, state_seq)); + cur = ggml_add(ctx0, cur, llm_build_time_mix_rwkv6(ctx0, layer, x_norm, x_prev, &wkv_states)); ggml_build_forward_expand(gf, cur); ggml_build_forward_expand( gf, @@ -15105,45 +15118,35 @@ struct llm_build_context { ggml_view_1d( ctx0, kv_self.v_l[layer_i], - hparams.n_embd_v_s() * n_kv, + hparams.n_embd_v_s() * n_seqs, hparams.n_embd_v_s() * kv_head * ggml_type_size(kv_self.v_l[layer_i]->type) ) ) ); + struct ggml_tensor * last_norm = ggml_view_3d(ctx0, x_norm, n_embd, 1, n_seqs, x_norm->nb[1], x_norm->nb[2], (n_seq_tokens-1)*n_embd*ggml_element_size(x_norm)); ggml_build_forward_expand( gf, ggml_cpy( - ctx0, - ggml_view_1d( - ctx0, - tmp, - n_embd * n_kv, - n_tokens * n_embd * ggml_type_size(kv_self.k_l[layer_i]->type) - ), - ggml_view_1d(ctx0, token_shift, n_embd * n_kv, 0) + ctx0, last_norm, + ggml_view_1d(ctx0, token_shift, n_embd * n_seqs, 0) ) ); x_norm = llm_build_norm(ctx0, cur, hparams, layer->attn_norm_2, layer->attn_norm_2_b, LLM_NORM, cb, layer_i); - tmp = ggml_rwkv_token_shift(ctx0, ffn_shift, x_norm, state_seq); - x_prev = ggml_reshape_2d( + x_prev = ggml_concat( ctx0, - ggml_view_1d(ctx0, tmp, n_embd * n_tokens, 0), - n_embd, n_tokens + ffn_shift, + ggml_view_3d(ctx0, x_norm, n_embd, n_seq_tokens - 1, n_seqs, x_norm->nb[1], x_norm->nb[2], 0), + 1 ); cur = ggml_add(ctx0, cur, llm_build_channel_mix_rwkv6(ctx0, layer, x_norm, x_prev)); + last_norm = ggml_view_3d(ctx0, x_norm, n_embd, 1, n_seqs, x_norm->nb[1], x_norm->nb[2], (n_seq_tokens-1)*n_embd*ggml_element_size(x_norm)); ggml_build_forward_expand(gf, cur); ggml_build_forward_expand( gf, ggml_cpy( - ctx0, - ggml_view_1d( - ctx0, - tmp, - n_embd * n_kv, - n_tokens * n_embd * ggml_type_size(kv_self.k_l[layer_i]->type) - ), - ggml_view_1d(ctx0, token_shift, n_embd * n_kv, n_kv * n_embd * ggml_type_size(kv_self.k_l[layer_i]->type)) + ctx0, last_norm, + ggml_view_1d(ctx0, token_shift, n_embd * n_seqs, n_embd * n_seqs * ggml_element_size(token_shift)) ) ); @@ -15151,7 +15154,7 @@ struct llm_build_context { ctx0, ggml_permute( ctx0, - ggml_reshape_3d(ctx0, token_shift, n_embd, n_kv, 2), + ggml_reshape_3d(ctx0, token_shift, n_embd, n_seqs, 2), 0, 2, 1, 3 ) ); @@ -15160,8 +15163,8 @@ struct llm_build_context { gf, ggml_cpy( ctx0, - ggml_view_1d(ctx0, token_shift, n_embd * n_kv * 2, 0), - ggml_view_1d(ctx0, kv_self.k_l[layer_i], hparams.n_embd_k_s() * n_kv, hparams.n_embd_k_s() * kv_head * ggml_type_size(kv_self.k_l[layer_i]->type)) + ggml_view_1d(ctx0, token_shift, n_embd * n_seqs * 2, 0), + ggml_view_1d(ctx0, kv_self.k_l[layer_i], hparams.n_embd_k_s() * n_seqs, hparams.n_embd_k_s() * kv_head * ggml_type_size(kv_self.k_l[layer_i]->type)) ) ); @@ -15171,6 +15174,7 @@ struct llm_build_context { } ggml_tensor * inp_out_ids = build_inp_out_ids(); + cur = ggml_reshape_2d(ctx0, cur, n_embd, n_tokens); cur = ggml_get_rows(ctx0, cur, inp_out_ids); cur = llm_build_norm(ctx0, cur, hparams, model.output_norm, model.output_norm_b, LLM_NORM, cb, -1);