use workspace to optimize kvcache attention

examples/python/llama_kvcache_inference.py

@@ -67,9 +67,7 @@ def replace_onnx_with_attention_op():
             tmap["/model/layers." + str(i) + "/self_attn/Add_1_output_0"],
             tmap["/model/layers." + str(i) + "/self_attn/Transpose_2_output_0"]]
         outputs = [
-            tmap["/model/layers." + str(i) + "/self_attn/MatMul_1_output_0"],
+            tmap["/model/layers." + str(i) + "/self_attn/MatMul_1_output_0"]]
-            tmap[graph.outputs[1+i*2].name],
-            tmap[graph.outputs[2+i*2].name]]
 
         inputs_added = [graph.inputs[1]]
         outputs_removed = []
@@ -77,6 +75,7 @@ def replace_onnx_with_attention_op():
         graph.replace_with_attention(
             inputs, outputs, inputs_added, outputs_removed)
         
+    graph.outputs = [tmap[graph.outputs[0].name]]
     graph.cleanup(True).toposort()
     onnx.save(gs.export_onnx(graph), ONNX_MODEL_PATH, save_as_external_data=True)
 

include/core/graph_handler.h

@@ -74,10 +74,9 @@ class GraphHandlerObj {
     Tensor squeeze(Tensor input, Tensor output, Shape axes);
     Tensor unsqueeze(Tensor input, Tensor output, Shape axes);
     Tensor concat(TensorVec inputs, Tensor output, int dim);
-    TensorVec attentionKVCache(Tensor input_k_cache, Tensor input_v_cache,
+    Tensor attentionKVCache(Tensor input_k_cache, Tensor input_v_cache,
                             Tensor input_q, Tensor input_k, Tensor input_v,
-                               Tensor position_id, Tensor output_matmul,
+                            Tensor position_id, Tensor output_matmul);
-                               Tensor output_k_cache, Tensor output_v_cache);
     TensorVec split(Tensor input, std::optional<TensorVec> outputs, int axis,
                     std::variant<int, vector<int>> numOrRatio);
     Tensor gather(Tensor data, Tensor indices, Tensor output, int axis);

include/cuda/cuda_attention_kvcache.h

@@ -1,4 +1,5 @@
 #pragma once
+#include "core/common.h"
 #include <cstdio>
 
 struct AttentionKVCacheMetadata {

include/operators/attention_kvcache.h

@@ -22,21 +22,18 @@ class AttentionKVCacheObj : public OperatorObj {
      * @param input_v The value input tensor.
      * @param position_id The positon id of the query,
      * @param output_matmul The query output tensor.
-     * @param output_k_cache The output k_cache tensor.
-     * @param output_v_cache The output v_cache tensor.
      */
     AttentionKVCacheObj(GraphObj *graph, Tensor input_k_cache,
                         Tensor input_v_cache, Tensor input_q, Tensor input_k,
                         Tensor input_v, Tensor position_id,
-                        Tensor output_matmul, Tensor output_k_cache,
+                        Tensor output_matmul);
-                        Tensor output_v_cache);
     OP_CLONE(AttentionKVCacheObj);
 
     optional<vector<Shape>> inferShape(const TensorVec &inputs) override;
 
     std::string toString() const override;
     int numInputs() const override { return 6; }
-    int numOutputs() const override { return 3; }
+    int numOutputs() const override { return 1; }
     int getDim() const { return dim; }
 
   private:

pyinfinitensor/src/pyinfinitensor/onnx.py

@@ -660,7 +660,7 @@ class OnnxStub:
                     next((attr.i for attr in node.attribute if attr.name == "axis")),
                 )
             elif node.op_type == "AttentionKVCache":
-                tensors[node.output[0]], tensors[node.output[1]], tensors[node.output[2]] = self.handler.attentionKVCache(
+                tensors[node.output[0]] = self.handler.attentionKVCache(
                     tensors[node.input[0]],
                     tensors[node.input[1]],
                     tensors[node.input[2]],
@@ -668,8 +668,6 @@ class OnnxStub:
                     tensors[node.input[4]],
                     tensors[node.input[5]],
                     tensors.get(node.output[0]),
-                    tensors.get(node.output[1]),
-                    tensors.get(node.output[2]),
                 )
             elif node.op_type == "Split":
                 split = (

src/core/graph_handler.cc

@@ -324,25 +324,24 @@ Tensor GraphHandlerObj::concat(TensorVec inputs, Tensor output, int dim) {
     }
 }
 
-TensorVec GraphHandlerObj::attentionKVCache(
+Tensor GraphHandlerObj::attentionKVCache(Tensor input_k_cache,
-    Tensor input_k_cache, Tensor input_v_cache, Tensor input_q, Tensor input_k,
+                                         Tensor input_v_cache, Tensor input_q,
-    Tensor input_v, Tensor position_id, Tensor output_matmul,
+                                         Tensor input_k, Tensor input_v,
-    Tensor output_k_cache, Tensor output_v_cache) {
+                                         Tensor position_id,
-    if (output_matmul && output_k_cache && output_v_cache) {
+                                         Tensor output_matmul) {
+    if (output_matmul) {
         g->addOpWithOutputs<AttentionKVCacheObj>(
             std::move(input_k_cache), std::move(input_v_cache),
             std::move(input_q), std::move(input_k), std::move(input_v),
-            std::move(position_id), output_matmul, output_k_cache,
+            std::move(position_id), output_matmul);
-            output_v_cache);
+        return output_matmul;
-        return {output_matmul, output_k_cache, output_v_cache};
     } else {
         return g
             ->addOp<AttentionKVCacheObj>(
                 std::move(input_k_cache), std::move(input_v_cache),
                 std::move(input_q), std::move(input_k), std::move(input_v),
-                std::move(position_id), output_matmul, output_k_cache,
+                std::move(position_id), output_matmul)
-                output_v_cache)
+            ->getOutput();
-            ->getOutputs();
     }
 }
 

src/core/operator.cc

@@ -67,10 +67,8 @@ bool OperatorObj::checkValid(GraphObj *graph) {
     if (graph) { // if graph != nullptr, outputs should be created
         auto dataTypes = inferDataType();
         for (size_t i = 0; i < outputs.size(); i++) {
-            if (!outputs[i])
+            IT_ASSERT(!outputs[i], "Find empty output while operator creation");
             outputs[i] = graph->addTensor(shapes[i], dataTypes[i]);
-            else if (shapes[i] != outputs[i]->getDims())
-                return false;
         }
     } else { // if outputs have been created, check their shapes
         for (size_t i = 0; i < shapes.size(); ++i) {

src/kernels/cuda/attention_kvcache.cc

@@ -21,8 +21,7 @@ class AttentionKVCacheCompute {
   public:
     void do_compute(Tensor input_k_cache, Tensor input_v_cache, Tensor input_q,
                     Tensor input_k, Tensor input_v, Tensor position_id,
-                    Tensor output_matmul, Tensor output_temp_O,
+                    Tensor output_matmul, CudaPtr p_workspace) const {
-                    Tensor output_temp_sum) const {
         AttentionKVCacheMetadata metadata;
         initAttentionKVCacheMetadata(metadata, input_v_cache);
 
@@ -33,9 +32,8 @@ class AttentionKVCacheCompute {
                                  input_v->getRawDataPtr<float *>(),
                                  position_id->getRawDataPtr<int *>(),
                                  output_matmul->getRawDataPtr<float *>(),
-                                 metadata,
+                                 metadata, (float *)p_workspace,
-                                 output_temp_O->getRawDataPtr<float *>(),
+                                 (float *)(p_workspace + (1ll << 30)));
-                                 output_temp_sum->getRawDataPtr<float *>());
     }
 };
 
@@ -44,10 +42,14 @@ class AttentionKVCacheCuda : private AttentionKVCacheCompute,
     void compute(const Operator &_op,
                  const RuntimeObj *_context) const override {
         IT_ASSERT(_op->getDType() == DataType::Float32);
-        do_compute(
+
-            _op->getInputs()[0], _op->getInputs()[1], _op->getInputs()[2],
+        size_t workspaceSize = 2ll << 30;
-            _op->getInputs()[3], _op->getInputs()[4], _op->getInputs()[5],
+        auto context = dynamic_cast<const CudaRuntimeObj *>(_context);
-            _op->getOutputs()[0], _op->getOutputs()[1], _op->getOutputs()[2]);
+        CudaPtr idxWsData = context->getWorkspace(workspaceSize);
+        do_compute(_op->getInputs()[0], _op->getInputs()[1],
+                   _op->getInputs()[2], _op->getInputs()[3],
+                   _op->getInputs()[4], _op->getInputs()[5],
+                   _op->getOutputs()[0], idxWsData);
     }
 };
 

src/operators/attention_kvcache.cc

@@ -2,14 +2,15 @@
 #include "utils/operator_utils.h"
 
 namespace infini {
-AttentionKVCacheObj::AttentionKVCacheObj(
+AttentionKVCacheObj::AttentionKVCacheObj(GraphObj *graph, Tensor input_k_cache,
-    GraphObj *graph, Tensor input_k_cache, Tensor input_v_cache, Tensor input_q,
+                                         Tensor input_v_cache, Tensor input_q,
-    Tensor input_k, Tensor input_v, Tensor position_id, Tensor output_matmul,
+                                         Tensor input_k, Tensor input_v,
-    Tensor output_k_cache, Tensor output_v_cache)
+                                         Tensor position_id,
+                                         Tensor output_matmul)
     : OperatorObj(OpType::AttentionKVCache,
                   TensorVec{input_k_cache, input_v_cache, input_q, input_k,
                             input_v, position_id},
-                  TensorVec{output_matmul, output_k_cache, output_v_cache}) {
+                  {output_matmul}) {
     int rank = inputs[0]->getRank();
     IT_ASSERT(rank == 4);
     dim = 2;
@@ -22,7 +23,7 @@ AttentionKVCacheObj::inferShape(const TensorVec &inputs) {
     Shape dims = inputs[0]->getDims();
     ShapeElem n = dims.at(dim);
     dims[dim] = n + 1;
-    return {{inputs[2]->getDims(), dims, dims}};
+    return {{inputs[2]->getDims()}};
 }
 
 std::string AttentionKVCacheObj::toString() const {

test/kernels/cuda/test_cuda_attention.cc

@@ -23,7 +23,7 @@ TEST(AttentionKVCache, Cuda) {
 
     auto op = gCuda->addOp<AttentionKVCacheObj>(
         input_k_cache_d, input_v_cache_d, input_q_d, input_k_d, input_v_d,
-        position_id_d, nullptr, nullptr, nullptr);
+        position_id_d, nullptr);
     gCuda->dataMalloc();
 
     input_q_d->setData(OneGenerator());