@fugood/llama.node 1.0.1 → 1.0.3

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

@@ -105,10 +105,30 @@ public:
         res.resize(n);
 
         for (uint32_t j = 0; j < n; ++j) {
-            res.pos[j] = pos[i + j];
-            res.seq[j] = seq[i + j];
+            const auto idx = i + j;
 
-            assert(shift[i + j] == 0);
+            res.pos[j] = pos[idx];
+            res.seq[j] = seq[idx];
+
+            assert(shift[idx] == 0);
+        }
+
+        return res;
+    }
+
+    // copy the state of cells [idxs[0], idxs[1], ..., idxs[idxs.size() - 1])
+    llama_kv_cells_unified cp(const std::vector<uint32_t> & idxs) const {
+        llama_kv_cells_unified res;
+
+        res.resize(idxs.size());
+
+        for (uint32_t j = 0; j < idxs.size(); ++j) {
+            const auto idx = idxs[j];
+
+            res.pos[j] = pos[idx];
+            res.seq[j] = seq[idx];
+
+            assert(shift[idx] == 0);
         }
 
         return res;
@@ -119,26 +139,58 @@ public:
         assert(i + other.pos.size() <= pos.size());
 
         for (uint32_t j = 0; j < other.pos.size(); ++j) {
-            if (pos[i + j] == -1 && other.pos[j] != -1) {
+            const auto idx = i + j;
+
+            if (pos[idx] == -1 && other.pos[j] != -1) {
                 used.insert(i + j);
             }
 
-            if (pos[i + j] != -1 && other.pos[j] == -1) {
+            if (pos[idx] != -1 && other.pos[j] == -1) {
                 used.erase(i + j);
             }
 
-            if (pos[i + j] != -1) {
+            if (pos[idx] != -1) {
                 seq_pos_rm(i + j);
             }
 
-            pos[i + j] = other.pos[j];
-            seq[i + j] = other.seq[j];
+            pos[idx] = other.pos[j];
+            seq[idx] = other.seq[j];
 
-            if (pos[i + j] != -1) {
+            if (pos[idx] != -1) {
                 seq_pos_add(i + j);
             }
 
-            assert(shift[i + j] == 0);
+            assert(shift[idx] == 0);
+        }
+    }
+
+    // set the state of cells [idxs[0], idxs[1], ..., idxs[idxs.size() - 1])
+    void set(const std::vector<uint32_t> & idxs, const llama_kv_cells_unified & other) {
+        assert(idxs.size() == other.pos.size());
+
+        for (uint32_t j = 0; j < other.pos.size(); ++j) {
+            const auto idx = idxs[j];
+
+            if (pos[idx] == -1 && other.pos[j] != -1) {
+                used.insert(idx);
+            }
+
+            if (pos[idx] != -1 && other.pos[j] == -1) {
+                used.erase(idx);
+            }
+
+            if (pos[idx] != -1) {
+                seq_pos_rm(idx);
+            }
+
+            pos[idx] = other.pos[j];
+            seq[idx] = other.seq[j];
+
+            if (pos[idx] != -1) {
+                seq_pos_add(idx);
+            }
+
+            assert(shift[idx] == 0);
         }
     }
 

package/src/llama.cpp/src/llama-memory-hybrid.cpp

@@ -70,7 +70,7 @@ llama_memory_context_ptr llama_memory_hybrid::init_batch(llama_batch_allocr & ba
                 // if all tokens are output, split by sequence
                 ubatch = balloc.split_seq(n_ubatch);
             } else {
-                ubatch = balloc.split_equal(n_ubatch);
+                ubatch = balloc.split_equal(n_ubatch, false);
             }
 
             if (ubatch.n_tokens == 0) {
@@ -80,6 +80,11 @@ llama_memory_context_ptr llama_memory_hybrid::init_batch(llama_batch_allocr & ba
             ubatches.push_back(std::move(ubatch)); // NOLINT
         }
 
+        if (balloc.get_n_used() < balloc.get_n_tokens()) {
+            // failed to find a suitable split
+            break;
+        }
+
         // prepare the recurrent batches first
         if (!mem_recr->prepare(ubatches)) {
             // TODO: will the recurrent cache be in an undefined context at this point?
@@ -195,11 +200,11 @@ llama_memory_hybrid_context::llama_memory_hybrid_context(
 
 llama_memory_hybrid_context::llama_memory_hybrid_context(
               llama_memory_hybrid * mem,
-            std::vector<uint32_t>   heads_attn,
+                  slot_info_vec_t   sinfos_attn,
         std::vector<llama_ubatch>   ubatches) :
     ubatches(std::move(ubatches)),
     // note: here we copy the ubatches. not sure if this is ideal
-    ctx_attn(new llama_kv_cache_unified_context(mem->get_mem_attn(), std::move(heads_attn), this->ubatches)),
+    ctx_attn(new llama_kv_cache_unified_context(mem->get_mem_attn(), std::move(sinfos_attn), this->ubatches)),
     ctx_recr(new llama_memory_recurrent_context(mem->get_mem_recr(),                        this->ubatches)),
     status(llama_memory_status_combine(ctx_attn->get_status(), ctx_recr->get_status())) {
 }
@@ -218,7 +223,7 @@ bool llama_memory_hybrid_context::next() {
 }
 
 bool llama_memory_hybrid_context::apply() {
-    assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
+    assert(!llama_memory_status_is_fail(status));
 
     bool res = true;
 

package/src/llama.cpp/src/llama-memory-hybrid.h

@@ -92,6 +92,8 @@ private:
 
 class llama_memory_hybrid_context : public llama_memory_context_i {
 public:
+    using slot_info_vec_t = llama_kv_cache_unified::slot_info_vec_t;
+
     // init failure
     explicit llama_memory_hybrid_context(llama_memory_status status);
 
@@ -107,7 +109,7 @@ public:
     // init success
     llama_memory_hybrid_context(
               llama_memory_hybrid * mem,
-            std::vector<uint32_t>   heads_attn,
+                  slot_info_vec_t   sinfos_attn,
         std::vector<llama_ubatch>   ubatches);
 
     ~llama_memory_hybrid_context() = default;

package/src/llama.cpp/src/llama-memory-recurrent.cpp

@@ -363,30 +363,40 @@ llama_pos llama_memory_recurrent::seq_pos_max(llama_seq_id seq_id) const {
 }
 
 llama_memory_context_ptr llama_memory_recurrent::init_batch(llama_batch_allocr & balloc, uint32_t n_ubatch, bool embd_all) {
-    std::vector<llama_ubatch> ubatches;
+    do {
+        balloc.split_reset();
 
-    while (true) {
-        llama_ubatch ubatch;
+        std::vector<llama_ubatch> ubatches;
+        while (true) {
+            llama_ubatch ubatch;
 
-        if (embd_all) {
-            // if all tokens are output, split by sequence
-            ubatch = balloc.split_seq(n_ubatch);
-        } else {
-            ubatch = balloc.split_equal(n_ubatch);
+            if (embd_all) {
+                // if all tokens are output, split by sequence
+                ubatch = balloc.split_seq(n_ubatch);
+            } else {
+                ubatch = balloc.split_equal(n_ubatch, false);
+            }
+
+            if (ubatch.n_tokens == 0) {
+                break;
+            }
+
+            ubatches.push_back(std::move(ubatch)); // NOLINT
         }
 
-        if (ubatch.n_tokens == 0) {
+        if (balloc.get_n_used() < balloc.get_n_tokens()) {
+            // failed to find a suitable split
             break;
         }
 
-        ubatches.push_back(std::move(ubatch)); // NOLINT
-    }
+        if (!prepare(ubatches)) {
+            break;
+        }
 
-    if (!prepare(ubatches)) {
-        return std::make_unique<llama_memory_recurrent_context>(LLAMA_MEMORY_STATUS_FAILED_PREPARE);
-    }
+        return std::make_unique<llama_memory_recurrent_context>(this, std::move(ubatches));
+    } while (false);
 
-    return std::make_unique<llama_memory_recurrent_context>(this, std::move(ubatches));
+    return std::make_unique<llama_memory_recurrent_context>(LLAMA_MEMORY_STATUS_FAILED_PREPARE);
 }
 
 llama_memory_context_ptr llama_memory_recurrent::init_full() {
@@ -1066,7 +1076,15 @@ bool llama_memory_recurrent_context::next() {
 }
 
 bool llama_memory_recurrent_context::apply() {
-    assert(status == LLAMA_MEMORY_STATUS_SUCCESS);
+    assert(!llama_memory_status_is_fail(status));
+
+    // no ubatches -> this is an update
+    if (ubatches.empty()) {
+        // recurrent cache never performs updates
+        assert(status == LLAMA_MEMORY_STATUS_NO_UPDATE);
+
+        return true;
+    }
 
     mem->find_slot(ubatches[i_next]);
 

package/src/llama.cpp/src/llama-memory.cpp

@@ -40,3 +40,20 @@ llama_memory_status llama_memory_status_combine(llama_memory_status s0, llama_me
     // if either status has an update, then the combined status has an update
     return has_update ? LLAMA_MEMORY_STATUS_SUCCESS : LLAMA_MEMORY_STATUS_NO_UPDATE;
 }
+
+bool llama_memory_status_is_fail(llama_memory_status status) {
+    switch (status) {
+        case LLAMA_MEMORY_STATUS_SUCCESS:
+        case LLAMA_MEMORY_STATUS_NO_UPDATE:
+            {
+                return false;
+            }
+        case LLAMA_MEMORY_STATUS_FAILED_PREPARE:
+        case LLAMA_MEMORY_STATUS_FAILED_COMPUTE:
+            {
+                return true;
+            }
+    }
+
+    return false;
+}

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

@@ -31,6 +31,9 @@ enum llama_memory_status {
 // useful for implementing hybrid memory types (e.g. iSWA)
 llama_memory_status llama_memory_status_combine(llama_memory_status s0, llama_memory_status s1);
 
+// helper function for checking if a memory status indicates a failure
+bool llama_memory_status_is_fail(llama_memory_status status);
+
 // the interface for managing the memory context during batch processing
 // this interface is implemented per memory type. see:
 //   - llama_kv_cache_unified_context