@fugood/llama.node 1.1.10 → 1.1.11

package/src/llama.cpp/src/llama-context.cpp

@@ -41,7 +41,6 @@ llama_context::llama_context(
     cparams.yarn_beta_slow   = params.yarn_beta_slow;
     cparams.embeddings       = params.embeddings;
     cparams.offload_kqv      = params.offload_kqv;
-    cparams.flash_attn       = params.flash_attn;
     cparams.no_perf          = params.no_perf;
     cparams.pooling_type     = params.pooling_type;
     cparams.warmup           = false;
@@ -86,6 +85,8 @@ llama_context::llama_context(
         cparams.causal_attn = params.attention_type == LLAMA_ATTENTION_TYPE_CAUSAL;
     }
 
+    cparams.flash_attn = params.flash_attn_type != LLAMA_FLASH_ATTN_TYPE_DISABLED;
+
     // with causal attention, the batch size is limited by the context size
     cparams.n_batch = cparams.causal_attn ? std::min(cparams.n_ctx, params.n_batch) : params.n_batch;
 
@@ -102,16 +103,6 @@ llama_context::llama_context(
     cparams.op_offload = params.op_offload;
     cparams.kv_unified = params.kv_unified;
 
-    {
-        const char * LLAMA_SET_ROWS = getenv("LLAMA_SET_ROWS");
-        supports_set_rows = LLAMA_SET_ROWS ? (atoi(LLAMA_SET_ROWS) != 0) : supports_set_rows;
-
-        if (!supports_set_rows && !cparams.kv_unified) {
-            LLAMA_LOG_WARN("%s: non-unified KV cache requires ggml_set_rows() - forcing unified KV cache\n", __func__);
-            cparams.kv_unified = true;
-        }
-    }
-
     {
         const char * LLAMA_GRAPH_REUSE_DISABLE = getenv("LLAMA_GRAPH_REUSE_DISABLE");
         graph_reuse_disable = LLAMA_GRAPH_REUSE_DISABLE ? (atoi(LLAMA_GRAPH_REUSE_DISABLE) != 0) : graph_reuse_disable;
@@ -129,7 +120,7 @@ llama_context::llama_context(
     LLAMA_LOG_INFO("%s: n_batch       = %u\n",   __func__, cparams.n_batch);
     LLAMA_LOG_INFO("%s: n_ubatch      = %u\n",   __func__, cparams.n_ubatch);
     LLAMA_LOG_INFO("%s: causal_attn   = %d\n",   __func__, cparams.causal_attn);
-    LLAMA_LOG_INFO("%s: flash_attn    = %d\n",   __func__, cparams.flash_attn);
+    LLAMA_LOG_INFO("%s: flash_attn    = %s\n",   __func__, llama_flash_attn_type_name(params.flash_attn_type));
     LLAMA_LOG_INFO("%s: kv_unified    = %s\n",   __func__, cparams.kv_unified ? "true" : "false");
     LLAMA_LOG_INFO("%s: freq_base     = %.1f\n", __func__, cparams.rope_freq_base);
     LLAMA_LOG_INFO("%s: freq_scale    = %g\n",   __func__, cparams.rope_freq_scale);
@@ -279,8 +270,8 @@ llama_context::llama_context(
         }
     }
 
-    // reserve worst-case graph
-    if (!hparams.vocab_only && memory) {
+    // resolve automatic Flash Attention use and reserve worst-case graph
+    if (!hparams.vocab_only) {
         const uint32_t n_seqs = cparams.kv_unified ? 1 : cparams.n_seq_max;
         const uint32_t n_tokens = std::min(cparams.n_ctx, cparams.n_ubatch);
 
@@ -292,11 +283,13 @@ llama_context::llama_context(
         int n_splits_tg = -1;
         int n_nodes_tg  = -1;
 
-        // simulate full KV cache
-
-        const auto mctx = memory->init_full();
-        if (!mctx) {
-            throw std::runtime_error("failed to initialize KV cache");
+        llama_memory_context_ptr mctx;
+        if (memory) {
+            LLAMA_LOG_DEBUG("%s: reserving full memory module\n", __func__);
+            mctx = memory->init_full();
+            if (!mctx) {
+                throw std::runtime_error("failed to initialize memory module");
+            }
         }
 
         cross.v_embd.clear();
@@ -308,6 +301,48 @@ llama_context::llama_context(
                 throw std::runtime_error("failed to allocate compute pp buffers");
             }
 
+            if (params.flash_attn_type == LLAMA_FLASH_ATTN_TYPE_AUTO) {
+                ggml_backend_sched_alloc_graph(sched.get(), gf);
+
+                const size_t prefix_len = strlen(LLAMA_TENSOR_NAME_FATTN) + 1;
+                bool fa_device_mismatch = false;
+                for (int i = 0; i < ggml_graph_n_nodes(gf); i++) {
+                    ggml_tensor * n = ggml_graph_node(gf, i);
+                    if (n->op != GGML_OP_FLASH_ATTN_EXT) {
+                        continue;
+                    }
+                    ggml_backend_dev_t device_fa = ggml_backend_get_device(
+                        ggml_backend_sched_get_tensor_backend(sched.get(), n));
+
+                    // TODO: instead of the tensor names, use a map to keep track of which (FA) tensors belong to which layer
+                    GGML_ASSERT(strncmp(n->name, LLAMA_TENSOR_NAME_FATTN "-", prefix_len) == 0);
+                    const int il = std::stoi(n->name + prefix_len);
+                    ggml_backend_dev_t device_kv = model.dev_layer(il);
+                    if (device_fa != device_kv) {
+                        LLAMA_LOG_WARN("%s: layer %d is assigned to device %s but the Flash Attention tensor "
+                            "is assigned to device %s (usually due to missing support)\n",
+                            __func__, il, ggml_backend_dev_name(device_kv), ggml_backend_dev_name(device_fa));
+                        // FIXME: fa_device_mismatch logic is wrong for --no-kv-offload, but this is broken anyways
+                        fa_device_mismatch = true;
+                        break;
+                    }
+                }
+                if (fa_device_mismatch) {
+                    cparams.flash_attn = false;
+                    LLAMA_LOG_WARN("%s: Flash Attention was auto, set to disabled\n", __func__);
+                    if (ggml_is_quantized(params.type_v)) {
+                        throw std::runtime_error("quantized V cache was requested, but this requires Flash Attention");
+                    }
+                    auto * gf = graph_reserve(n_tokens, n_seqs, n_tokens, mctx.get());
+                    if (!gf) {
+                        throw std::runtime_error("failed to allocate compute pp buffers");
+                    }
+                } else {
+                    cparams.flash_attn = true;
+                    LLAMA_LOG_INFO("%s: Flash Attention was auto, set to enabled\n", __func__);
+                }
+            }
+
             n_splits_pp = ggml_backend_sched_get_n_splits(sched.get());
             n_nodes_pp  = ggml_graph_n_nodes(gf);
         }
@@ -888,12 +923,6 @@ int llama_context::encode(const llama_batch & batch_inp) {
         }
     }
 
-    if (!supports_set_rows) {
-        // Reset state for the next token before backend sync, to allow the CPU activities in the reset to
-        // overlap with device computation.
-        ggml_backend_sched_reset(sched.get());
-    }
-
     // TODO: hacky solution
     if (model.arch == LLM_ARCH_T5 && t_embd) {
         //cross.t_embd = t_embd;
@@ -1056,7 +1085,7 @@ int llama_context::decode(const llama_batch & batch_inp) {
         const auto * res = process_ubatch(ubatch, LLM_GRAPH_TYPE_DECODER, mctx.get(), status);
 
         if (!res) {
-            // the last ubatch failed or was aborted -> remove all positions of that ubatch from the KV cache
+            // the last ubatch failed or was aborted -> remove all positions of that ubatch from the memory module
             llama_pos pos_min[LLAMA_MAX_SEQ];
             for (int s = 0; s < LLAMA_MAX_SEQ; ++s) {
                 pos_min[s] = std::numeric_limits<llama_pos>::max();
@@ -1073,7 +1102,7 @@ int llama_context::decode(const llama_batch & batch_inp) {
                     continue;
                 }
 
-                LLAMA_LOG_WARN("%s: removing KV cache entries for seq_id = %d, pos = [%d, +inf)\n", __func__, s, pos_min[s]);
+                LLAMA_LOG_WARN("%s: removing memory module entries for seq_id = %d, pos = [%d, +inf)\n", __func__, s, pos_min[s]);
 
                 memory->seq_rm(s, pos_min[s], -1);
             }
@@ -1224,12 +1253,6 @@ int llama_context::decode(const llama_batch & batch_inp) {
     // wait for the computation to finish (automatically done when obtaining the model output)
     //synchronize();
 
-    if (!supports_set_rows) {
-        // Reset state for the next token before backend sync, to allow the CPU activities in the reset to
-        // overlap with device computation.
-        ggml_backend_sched_reset(sched.get());
-    }
-
     return 0;
 }
 
@@ -1857,7 +1880,7 @@ size_t llama_context::state_write_data(llama_io_write_i & io) {
     }
 
     if (memory != nullptr) {
-        LLAMA_LOG_DEBUG("%s: - writing KV self\n", __func__);
+        LLAMA_LOG_DEBUG("%s: - writing memory module\n", __func__);
         memory->state_write(io);
     }
 
@@ -1943,7 +1966,7 @@ size_t llama_context::state_read_data(llama_io_read_i & io) {
     }
 
     if (memory) {
-        LLAMA_LOG_DEBUG("%s: - reading KV self\n", __func__);
+        LLAMA_LOG_DEBUG("%s: - reading memory module\n", __func__);
 
         memory->state_read(io);
     }
@@ -2228,6 +2251,7 @@ llama_context_params llama_context_default_params() {
         /*.rope_scaling_type           =*/ LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED,
         /*.pooling_type                =*/ LLAMA_POOLING_TYPE_UNSPECIFIED,
         /*.attention_type              =*/ LLAMA_ATTENTION_TYPE_UNSPECIFIED,
+        /*.flash_attn_type             =*/ LLAMA_FLASH_ATTN_TYPE_AUTO,
         /*.rope_freq_base              =*/ 0.0f,
         /*.rope_freq_scale             =*/ 0.0f,
         /*.yarn_ext_factor             =*/ -1.0f,
@@ -2244,7 +2268,6 @@ llama_context_params llama_context_default_params() {
         /*.abort_callback_data         =*/ nullptr,
         /*.embeddings                  =*/ false,
         /*.offload_kqv                 =*/ true,
-        /*.flash_attn                  =*/ false,
         /*.no_perf                     =*/ true,
         /*.op_offload                  =*/ true,
         /*.swa_full                    =*/ true,
@@ -2272,12 +2295,30 @@ llama_context * llama_init_from_model(
         return nullptr;
     }
 
-    if (params.flash_attn && model->arch == LLM_ARCH_GROK) {
+    if (params.flash_attn_type != LLAMA_FLASH_ATTN_TYPE_DISABLED && model->arch == LLM_ARCH_GROK) {
         LLAMA_LOG_WARN("%s: flash_attn is not compatible with Grok - forcing off\n", __func__);
-        params.flash_attn = false;
+        params.flash_attn_type = LLAMA_FLASH_ATTN_TYPE_DISABLED;
+    }
+
+    if (params.flash_attn_type == LLAMA_FLASH_ATTN_TYPE_AUTO && ggml_is_quantized(params.type_k)) {
+        const uint32_t blck_size = ggml_blck_size(params.type_k);
+        if (model->hparams.n_embd_head_k % blck_size != 0) {
+            LLAMA_LOG_ERROR("%s: K cache type %s with block size %u does not divide n_embd_head_k=%u\n",
+                __func__, ggml_type_name(params.type_k), blck_size, model->hparams.n_embd_head_k);
+            return nullptr;
+        }
+    }
+
+    if (params.flash_attn_type == LLAMA_FLASH_ATTN_TYPE_AUTO && ggml_is_quantized(params.type_v)) {
+        const uint32_t blck_size = ggml_blck_size(params.type_v);
+        if (model->hparams.n_embd_head_v % blck_size != 0) {
+            LLAMA_LOG_ERROR("%s: V cache type %s with block size %u does not divide n_embd_head_k=%u\n",
+                __func__, ggml_type_name(params.type_v), blck_size, model->hparams.n_embd_head_v);
+            return nullptr;
+        }
     }
 
-    if (ggml_is_quantized(params.type_v) && !params.flash_attn) {
+    if (ggml_is_quantized(params.type_v) && params.flash_attn_type == LLAMA_FLASH_ATTN_TYPE_DISABLED) {
         LLAMA_LOG_ERROR("%s: V cache quantization requires flash_attn\n", __func__);
         return nullptr;
     }

package/src/llama.cpp/src/llama-context.h

@@ -283,10 +283,6 @@ private:
 
     bool has_evaluated_once = false;
 
-    // env: LLAMA_SET_ROWS (temporary)
-    // ref: https://github.com/ggml-org/llama.cpp/pull/14285
-    bool supports_set_rows = true;
-
     // env: LLAMA_GRAPH_REUSE_DISABLE
     bool graph_reuse_disable = false;
 

package/src/llama.cpp/src/llama-graph.cpp

@@ -314,8 +314,6 @@ bool llm_graph_input_attn_kv::can_reuse(const llm_graph_params & params) {
     res &= self_kq_mask->ne[0] == mctx->get_n_kv();
     res &= self_kq_mask->ne[1] == GGML_PAD(params.ubatch.n_tokens, GGML_KQ_MASK_PAD);
 
-    res &= mctx->get_supports_set_rows(); // TODO: tmp
-
     return res;
 }
 
@@ -350,8 +348,6 @@ bool llm_graph_input_attn_kv_iswa::can_reuse(const llm_graph_params & params) {
     res &= self_kq_mask_swa->ne[0] == mctx->get_swa()->get_n_kv();
     res &= self_kq_mask_swa->ne[1] == GGML_PAD(params.ubatch.n_tokens, GGML_KQ_MASK_PAD);
 
-    res &= mctx->get_base()->get_supports_set_rows(); // TODO: tmp
-
     return res;
 }
 
@@ -1225,7 +1221,8 @@ ggml_tensor * llm_graph_context::build_attn_mha(
          ggml_tensor * kq_mask,
          ggml_tensor * sinks,
          ggml_tensor * v_mla,
-             float     kq_scale) const {
+               float   kq_scale,
+                 int   il) const {
     const bool v_trans = v->nb[1] > v->nb[2];
 
     // split the batch into streams if needed
@@ -1260,6 +1257,7 @@ ggml_tensor * llm_graph_context::build_attn_mha(
 
         cur = ggml_flash_attn_ext(ctx0, q, k, v, kq_mask, kq_scale, hparams.f_max_alibi_bias,
                                   hparams.attn_soft_cap ? hparams.f_attn_logit_softcapping : 0.0f);
+        cb(cur, LLAMA_TENSOR_NAME_FATTN, il);
 
         ggml_flash_attn_ext_add_sinks(cur, sinks);
         ggml_flash_attn_ext_set_prec (cur, GGML_PREC_F32);
@@ -1275,6 +1273,7 @@ ggml_tensor * llm_graph_context::build_attn_mha(
             // The permutations are noops and only change how the tensor data is interpreted.
             cur = ggml_permute(ctx0, cur, 0, 2, 1, 3);
             cur = ggml_mul_mat(ctx0, v_mla, cur);
+            cb(cur, "fattn_mla", il);
             cur = ggml_permute(ctx0, cur, 0, 2, 1, 3);
             cur = ggml_cont(ctx0, cur); // Needed because ggml_reshape_2d expects contiguous inputs.
 #endif
@@ -1283,6 +1282,7 @@ ggml_tensor * llm_graph_context::build_attn_mha(
         cur = ggml_reshape_2d(ctx0, cur, cur->ne[0]*cur->ne[1], cur->ne[2]*cur->ne[3]);
     } else {
         ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
+        cb(kq, "kq", il);
 
         // note: this op tends to require high floating point range
         //       while for some models F16 is enough, for others it is not, so we default to F32 here
@@ -1296,32 +1296,42 @@ ggml_tensor * llm_graph_context::build_attn_mha(
             // before the softmax below
 
             kq = ggml_tanh(ctx0, ggml_scale(ctx0, kq, 0.08838834764831845f/30.0f));
+            cb(kq, "kq_tanh", il);
             kq = ggml_scale(ctx0, kq, 30);
+            cb(kq, "kq_scaled", il);
         }
 
         if (hparams.attn_soft_cap) {
             kq = ggml_scale(ctx0, kq, 1.0f / hparams.f_attn_logit_softcapping);
+            cb(kq, "kq_scaled_1", il);
             kq = ggml_tanh (ctx0, kq);
+            cb(kq, "kq_tanh", il);
             kq = ggml_scale(ctx0, kq, hparams.f_attn_logit_softcapping);
+            cb(kq, "kq_scaled_2", il);
         }
 
         if (kq_b) {
             kq = ggml_add(ctx0, kq, kq_b);
+            cb(kq, "kq_plus_kq_b", il);
         }
 
         kq = ggml_soft_max_ext(ctx0, kq, kq_mask, kq_scale, hparams.f_max_alibi_bias);
         ggml_soft_max_add_sinks(kq, sinks);
+        cb(kq, "kq_soft_max", il);
 
         if (!v_trans) {
             // note: avoid this branch
             v = ggml_cont(ctx0, ggml_transpose(ctx0, v));
+            cb(v, "v_cont", il);
         }
 
         ggml_tensor * kqv = ggml_mul_mat(ctx0, v, kq);
+        cb(kqv, "kqv", il);
 
         // for MLA with the absorption optimization, we need to "decompress" from MQA back to MHA
         if (v_mla) {
             kqv = ggml_mul_mat(ctx0, v_mla, kqv);
+            cb(kqv, "kqv_mla", il);
         }
 
         cur = ggml_permute(ctx0, kqv, 0, 2, 1, 3);
@@ -1376,13 +1386,13 @@ ggml_tensor * llm_graph_context::build_attn(
 
     // [TAG_NO_CACHE_PAD]
     // TODO: if ubatch.equal_seqs() == true, we can split the three tensors below into ubatch.n_seqs_unq streams
-    assert(!ubatch.equal_seqs());
+    assert(!ubatch.equal_seqs() || (k_cur->ne[3] == 1 && k_cur->ne[3] == ubatch.n_seqs_unq));
 
     ggml_tensor * q = q_cur;
     ggml_tensor * k = k_cur;
     ggml_tensor * v = v_cur;
 
-    ggml_tensor * cur = build_attn_mha(q, k, v, kq_b, kq_mask, sinks, v_mla, kq_scale);
+    ggml_tensor * cur = build_attn_mha(q, k, v, kq_b, kq_mask, sinks, v_mla, kq_scale, il);
     cb(cur, "kqv_out", il);
 
     if (wo) {
@@ -1471,7 +1481,7 @@ ggml_tensor * llm_graph_context::build_attn(
     ggml_tensor * k = mctx_cur->get_k(ctx0, il);
     ggml_tensor * v = mctx_cur->get_v(ctx0, il);
 
-    ggml_tensor * cur = build_attn_mha(q, k, v, kq_b, kq_mask, sinks, v_mla, kq_scale);
+    ggml_tensor * cur = build_attn_mha(q, k, v, kq_b, kq_mask, sinks, v_mla, kq_scale, il);
     cb(cur, "kqv_out", il);
 
     if (wo) {
@@ -1538,7 +1548,7 @@ ggml_tensor * llm_graph_context::build_attn(
     ggml_tensor * k = mctx_cur->get_k(ctx0, il);
     ggml_tensor * v = mctx_cur->get_v(ctx0, il);
 
-    ggml_tensor * cur = build_attn_mha(q, k, v, kq_b, kq_mask, sinks, v_mla, kq_scale);
+    ggml_tensor * cur = build_attn_mha(q, k, v, kq_b, kq_mask, sinks, v_mla, kq_scale, il);
     cb(cur, "kqv_out", il);
 
     if (wo) {
@@ -1593,7 +1603,7 @@ ggml_tensor * llm_graph_context::build_attn(
     ggml_tensor * k = k_cur;
     ggml_tensor * v = v_cur;
 
-    ggml_tensor * cur = build_attn_mha(q, k, v, kq_b, kq_mask, sinks, v_mla, kq_scale);
+    ggml_tensor * cur = build_attn_mha(q, k, v, kq_b, kq_mask, sinks, v_mla, kq_scale, il);
     cb(cur, "kqv_out", il);
 
     if (wo) {

package/src/llama.cpp/src/llama-graph.h

@@ -687,7 +687,8 @@ struct llm_graph_context {
             ggml_tensor * kq_mask,
             ggml_tensor * sinks,   // [n_head_q]
             ggml_tensor * v_mla,   // [n_embd_head_v_mla, n_embd_head_v, n_head_v]
-                  float   kq_scale) const;
+                  float   kq_scale,
+                    int   il) const;
 
     llm_graph_input_attn_no_cache * build_attn_inp_no_cache() const;
 

package/src/llama.cpp/src/llama-impl.h

@@ -59,3 +59,5 @@ std::string llama_format_tensor_shape(const std::vector<int64_t> & ne);
 std::string llama_format_tensor_shape(const struct ggml_tensor * t);
 
 std::string gguf_kv_to_str(const struct gguf_context * ctx_gguf, int i);
+
+#define LLAMA_TENSOR_NAME_FATTN "__fattn__"

package/src/llama.cpp/src/llama-kv-cache.cpp

@@ -197,18 +197,6 @@ llama_kv_cache::llama_kv_cache(
 
     const char * LLAMA_KV_CACHE_DEBUG = getenv("LLAMA_KV_CACHE_DEBUG");
     debug = LLAMA_KV_CACHE_DEBUG ? atoi(LLAMA_KV_CACHE_DEBUG) : 0;
-
-    const char * LLAMA_SET_ROWS = getenv("LLAMA_SET_ROWS");
-    supports_set_rows = LLAMA_SET_ROWS ? atoi(LLAMA_SET_ROWS) != 0 : supports_set_rows;
-
-    if (!supports_set_rows) {
-        // ref: https://github.com/ggml-org/llama.cpp/pull/14363
-        GGML_ASSERT(unified && "cannot use non-unified KV cache without ggml_set_rows() support");
-    }
-
-    if (!supports_set_rows) {
-        LLAMA_LOG_WARN("%s: LLAMA_SET_ROWS=0, using old ggml_cpy() method for backwards compatibility\n", __func__);
-    }
 }
 
 void llama_kv_cache::clear(bool data) {
@@ -551,11 +539,8 @@ llama_kv_cache::slot_info_vec_t llama_kv_cache::prepare(const std::vector<llama_
     bool success = true;
 
     for (const auto & ubatch : ubatches) {
-        // non-continuous slots require support for ggml_set_rows()
-        const bool cont = supports_set_rows ? false : true;
-
         // only find a suitable slot for the ubatch. don't modify the cells yet
-        const auto sinfo_new = find_slot(ubatch, cont);
+        const auto sinfo_new = find_slot(ubatch, false);
         if (sinfo_new.empty()) {
             success = false;
             break;
@@ -771,8 +756,8 @@ llama_kv_cache::slot_info llama_kv_cache::find_slot(const llama_ubatch & ubatch,
             GGML_ASSERT(ubatch.seq_id  [s*n_tokens][0] == seq_id);
         }
 
-        res.s0 = std::min<llama_seq_id>(res.s0, seq_to_stream[seq_id]);
-        res.s1 = std::max<llama_seq_id>(res.s1, seq_to_stream[seq_id]);
+        res.s0 = std::min<uint32_t>(res.s0, seq_to_stream[seq_id]);
+        res.s1 = std::max<uint32_t>(res.s1, seq_to_stream[seq_id]);
 
         res.strm[s] = seq_to_stream[seq_id];
         res.idxs[s].reserve(n_tokens);
@@ -964,11 +949,11 @@ bool llama_kv_cache::get_has_shift() const {
     return result;
 }
 
-uint32_t llama_kv_cache::get_n_kv() const {
+uint32_t llama_kv_cache::get_n_kv(const slot_info & sinfo) const {
     uint32_t result = 0;
 
-    for (uint32_t s = 0; s < n_stream; ++s) {
-        const auto & cells = v_cells[s];
+    for (uint32_t s = 0; s < sinfo.n_stream(); ++s) {
+        const auto & cells = v_cells[sinfo.strm[s]];
 
         result = std::max(std::min(cells.size(), std::max(n_pad, GGML_PAD(cells.used_max_p1(), n_pad))), result);
     }
@@ -976,10 +961,6 @@ uint32_t llama_kv_cache::get_n_kv() const {
     return result;
 }
 
-bool llama_kv_cache::get_supports_set_rows() const {
-    return supports_set_rows;
-}
-
 ggml_tensor * llama_kv_cache::get_k(ggml_context * ctx, int32_t il, uint32_t n_kv, const slot_info & sinfo) const {
     const int32_t ikv = map_layer_ids.at(il);
 
@@ -1017,52 +998,42 @@ ggml_tensor * llama_kv_cache::get_v(ggml_context * ctx, int32_t il, uint32_t n_k
         // note: v->nb[1] <= v->nb[2]
         return ggml_view_4d(ctx, v,
                 hparams.n_embd_head_v, hparams.n_head_kv(il), n_kv, ns,
-                ggml_row_size(v->type, hparams.n_embd_head_v),            // v->nb[1]
-                ggml_row_size(v->type, n_embd_v_gqa),         // v->nb[2]
-                ggml_row_size(v->type, n_embd_v_gqa*kv_size), // v->nb[3]
+                ggml_row_size(v->type, hparams.n_embd_head_v),          // v->nb[1]
+                ggml_row_size(v->type, n_embd_v_gqa),                   // v->nb[2]
+                ggml_row_size(v->type, n_embd_v_gqa*kv_size),           // v->nb[3]
                 ggml_row_size(v->type, n_embd_v_gqa*kv_size)*sinfo.s0);
     }
 
     // note: v->nb[1] > v->nb[2]
     return ggml_view_4d(ctx, v,
             n_kv, hparams.n_head_kv(il), hparams.n_embd_head_v, ns,
-            ggml_row_size(v->type, kv_size*hparams.n_embd_head_v),    // v->nb[1]
-            ggml_row_size(v->type, kv_size),                          // v->nb[2]
-            ggml_row_size(v->type, kv_size*n_embd_v_gqa), // v->nb[3]
+            ggml_row_size(v->type, kv_size*hparams.n_embd_head_v),  // v->nb[1]
+            ggml_row_size(v->type, kv_size),                        // v->nb[2]
+            ggml_row_size(v->type, kv_size*n_embd_v_gqa),           // v->nb[3]
             ggml_row_size(v->type, kv_size*n_embd_v_gqa)*sinfo.s0);
 }
 
 ggml_tensor * llama_kv_cache::cpy_k(ggml_context * ctx, ggml_tensor * k_cur, ggml_tensor * k_idxs, int32_t il, const slot_info & sinfo) const {
+    GGML_UNUSED(sinfo);
+
     const int32_t ikv = map_layer_ids.at(il);
 
     auto * k = layers[ikv].k;
 
-    const int64_t n_embd_k_gqa = k->ne[0];
     const int64_t n_tokens = k_cur->ne[2];
 
     k_cur = ggml_reshape_2d(ctx, k_cur, k->ne[0], n_tokens);
 
-    if (k_idxs && supports_set_rows) {
-        if (k->ne[2] > 1) {
-            k = ggml_reshape_2d(ctx, k, k->ne[0], k->ne[1]*k->ne[2]);
-        }
-
-        return ggml_set_rows(ctx, k, k_cur, k_idxs);
+    if (k->ne[2] > 1) {
+        k = ggml_reshape_2d(ctx, k, k->ne[0], k->ne[1]*k->ne[2]);
     }
 
-    // TODO: fallback to old ggml_cpy() method for backwards compatibility
-    //       will be removed when ggml_set_rows() is adopted by all backends
-
-    GGML_ASSERT(n_stream == 1 && "n_stream > 1 not supported without LLAMA_SET_ROWS");
-
-    ggml_tensor * k_view = ggml_view_1d(ctx, k,
-            n_tokens*n_embd_k_gqa,
-            ggml_row_size(k->type, n_embd_k_gqa)*sinfo.head());
-
-    return ggml_cpy(ctx, k_cur, k_view);
+    return ggml_set_rows(ctx, k, k_cur, k_idxs);
 }
 
 ggml_tensor * llama_kv_cache::cpy_v(ggml_context * ctx, ggml_tensor * v_cur, ggml_tensor * v_idxs, int32_t il, const slot_info & sinfo) const {
+    GGML_UNUSED(sinfo);
+
     const int32_t ikv = map_layer_ids.at(il);
 
     auto * v = layers[ikv].v;
@@ -1072,48 +1043,25 @@ ggml_tensor * llama_kv_cache::cpy_v(ggml_context * ctx, ggml_tensor * v_cur, ggm
 
     v_cur = ggml_reshape_2d(ctx, v_cur, n_embd_v_gqa, n_tokens);
 
-    if (v_idxs && supports_set_rows) {
-        if (!v_trans) {
-            if (v->ne[2] > 1) {
-                v = ggml_reshape_2d(ctx, v, v->ne[0], v->ne[1]*v->ne[2]);
-            }
-
-            return ggml_set_rows(ctx, v, v_cur, v_idxs);
-        }
-
-        // [TAG_V_CACHE_VARIABLE]
-        if (n_embd_v_gqa < v->ne[0]) {
-            v_cur = ggml_pad(ctx, v_cur, v->ne[0] - n_embd_v_gqa, 0, 0, 0);
+    if (!v_trans) {
+        if (v->ne[2] > 1) {
+            v = ggml_reshape_2d(ctx, v, v->ne[0], v->ne[1]*v->ne[2]);
         }
 
-        // the row becomes a single element
-        ggml_tensor * v_view = ggml_reshape_2d(ctx, v, 1, v->ne[0]*v->ne[1]*v->ne[2]);
-
-        v_cur = ggml_reshape_2d(ctx, v_cur, 1, v_cur->ne[0]*v_cur->ne[1]);
-
-        return ggml_set_rows(ctx, v_view, v_cur, v_idxs);
+        return ggml_set_rows(ctx, v, v_cur, v_idxs);
     }
 
-    // TODO: fallback to old ggml_cpy() method for backwards compatibility
-    //       will be removed when ggml_set_rows() is adopted by all backends
-
-    GGML_ASSERT(n_stream == 1 && "n_stream > 1 not supported without LLAMA_SET_ROWS");
+    // [TAG_V_CACHE_VARIABLE]
+    if (n_embd_v_gqa < v->ne[0]) {
+        v_cur = ggml_pad(ctx, v_cur, v->ne[0] - n_embd_v_gqa, 0, 0, 0);
+    }
 
-    ggml_tensor * v_view = nullptr;
+    // the row becomes a single element
+    ggml_tensor * v_view = ggml_reshape_2d(ctx, v, 1, v->ne[0]*v->ne[1]*v->ne[2]);
 
-    if (!v_trans) {
-        v_view = ggml_view_1d(ctx, v,
-                n_tokens*n_embd_v_gqa,
-                ggml_row_size(v->type, n_embd_v_gqa)*sinfo.head());
-    } else {
-        v_cur = ggml_transpose(ctx, v_cur);
+    v_cur = ggml_reshape_2d(ctx, v_cur, 1, v_cur->ne[0]*v_cur->ne[1]);
 
-        v_view = ggml_view_2d(ctx, v, n_tokens, n_embd_v_gqa,
-                (v->ne[1]    )*ggml_element_size(v),
-                (sinfo.head())*ggml_element_size(v));
-    }
-
-    return ggml_cpy(ctx, v_cur, v_view);
+    return ggml_set_rows(ctx, v_view, v_cur, v_idxs);
 }
 
 ggml_tensor * llama_kv_cache::build_input_k_idxs(ggml_context * ctx, const llama_ubatch & ubatch) const {
@@ -1143,10 +1091,6 @@ ggml_tensor * llama_kv_cache::build_input_v_idxs(ggml_context * ctx, const llama
 }
 
 void llama_kv_cache::set_input_k_idxs(ggml_tensor * dst, const llama_ubatch * ubatch, const slot_info & sinfo) const {
-    if (!supports_set_rows) {
-        return;
-    }
-
     const uint32_t n_tokens = ubatch->n_tokens;
     GGML_ASSERT(n_tokens == (int64_t) sinfo.size()*sinfo.n_stream());
 
@@ -1163,10 +1107,6 @@ void llama_kv_cache::set_input_k_idxs(ggml_tensor * dst, const llama_ubatch * ub
 }
 
 void llama_kv_cache::set_input_v_idxs(ggml_tensor * dst, const llama_ubatch * ubatch, const slot_info & sinfo) const {
-    if (!supports_set_rows) {
-        return;
-    }
-
     const uint32_t n_tokens = ubatch->n_tokens;
     GGML_ASSERT(n_tokens == (int64_t) sinfo.size()*sinfo.n_stream());
 
@@ -1985,8 +1925,7 @@ bool llama_kv_cache_context::apply() {
     }
 
     kv->apply_ubatch(sinfos[i_cur], ubatches[i_cur]);
-
-    n_kv = kv->get_n_kv();
+    n_kv = kv->get_n_kv(sinfos[i_cur]);
 
     return true;
 }
@@ -2005,10 +1944,6 @@ uint32_t llama_kv_cache_context::get_n_kv() const {
     return n_kv;
 }
 
-bool llama_kv_cache_context::get_supports_set_rows() const {
-    return kv->get_supports_set_rows();
-}
-
 ggml_tensor * llama_kv_cache_context::get_k(ggml_context * ctx, int32_t il) const {
     return kv->get_k(ctx, il, n_kv, sinfos[i_cur]);
 }

package/src/llama.cpp/src/llama-kv-cache.h

@@ -38,8 +38,8 @@ public:
         using idx_vec_t = std::vector<uint32_t>;
 
         // number of streams: ns = s1 - s0 + 1
-        llama_seq_id s0;
-        llama_seq_id s1;
+        uint32_t s0;
+        uint32_t s1;
 
         std::vector<llama_seq_id> strm; // [ns]
         std::vector<idx_vec_t>    idxs; // [ns]
@@ -139,10 +139,7 @@ public:
     // graph_build API
     //
 
-    uint32_t get_n_kv() const;
-
-    // TODO: temporary
-    bool get_supports_set_rows() const;
+    uint32_t get_n_kv(const slot_info & sinfo) const;
 
     // get views of the current state of the cache
     ggml_tensor * get_k(ggml_context * ctx, int32_t il, uint32_t n_kv, const slot_info & sinfo) const;
@@ -215,10 +212,6 @@ private:
     // env: LLAMA_KV_CACHE_DEBUG
     int debug = 0;
 
-    // env: LLAMA_SET_ROWS (temporary)
-    // ref: https://github.com/ggml-org/llama.cpp/pull/14285
-    bool supports_set_rows = true;
-
     const llama_swa_type swa_type = LLAMA_SWA_TYPE_NONE;
 
     std::vector<ggml_context_ptr>        ctxs;
@@ -318,9 +311,6 @@ public:
 
     uint32_t get_n_kv() const;
 
-    // TODO: temporary
-    bool get_supports_set_rows() const;
-
     // get views of the current state of the cache
     ggml_tensor * get_k(ggml_context * ctx, int32_t il) const;
     ggml_tensor * get_v(ggml_context * ctx, int32_t il) const;