String tensor support (#832)

* Support string tensor in tensor set

* fixed bug in creating scalar tensor (with 0 dims)

* fix rdb v4 tests + allow string values not to be null-terminated when using tensor set from values

* PR fixes + testing string tensor concatenation and slicing via LLAPI, remove deep copy (not used)

Author: alonre24

.circleci/config.yml

@@ -112,7 +112,7 @@ commands:
       - save_cache:
           paths:
             - deps
-          key: build-dependencies-{{ checksum "get_deps.sh" }}
+          key: v1-dependencies-{{ checksum "get_deps.sh" }}
       - run:
           name: Build
           command: make -C opt all SHOW=1
@@ -151,7 +151,7 @@ commands:
       - checkout-all
       - restore_cache:
           keys:
-          - build-dependencies-{{ checksum "get_deps.sh" }}
+          - v1-dependencies-{{ checksum "get_deps.sh" }}
           # If no exact match is found will get dependencies from source
       - setup-build-system
       - run:
@@ -260,7 +260,7 @@ jobs:
       - checkout-all
       - restore_cache:
           keys:
-          - build-dependencies-{{ checksum "get_deps.sh" }}
+          - v1-dependencies-{{ checksum "get_deps.sh" }}
           # If no exact match is found will get dependencies from source
       - setup-build-system
       - run:
@@ -294,7 +294,7 @@ jobs:
       - checkout-all
       - restore_cache:
           keys:
-          - build-dependencies-{{ checksum "get_deps.sh" }}
+          - v1-dependencies-{{ checksum "get_deps.sh" }}
           # If no exact match is found will get dependencies from source
       - setup-build-system
       - run:

docs/commands.md

@@ -27,7 +27,7 @@ AI.TENSORSET <key> <type>
 _Arguments_
 
 * **key**: the tensor's key name
-* **type**: the tensor's data type can be one of: `FLOAT`, `DOUBLE`, `INT8`, `INT16`, `INT32`, `INT64`, `UINT8` or `UINT16`
+* **type**: the tensor's data type can be one of: `FLOAT`, `DOUBLE`, `INT8`, `INT16`, `INT32`, `INT64`, `UINT8`, `UINT16`, `BOOL` or `STRING`
 * **shape**: one or more dimensions, or the number of elements per axis, for the tensor
 * **BLOB**: indicates that data is in binary format and is provided via the subsequent `data` argument
 * **VALUES**: indicates that data is numeric and is provided by one or more subsequent `val` arguments
@@ -45,6 +45,8 @@ This will set the key 'mytensor' to the 2x2 RedisAI tensor:
 ```
 redis> AI.TENSORSET mytensor FLOAT 2 2 VALUES 1 2 3 4
 OK
+redis> AI.TENSORSET mytensor STRING 1 2 VALUES first second
+OK
 ```
 
 !!! note "Uninitialized Tensor Values"

src/CMakeLists.txt

@@ -34,6 +34,7 @@ ADD_LIBRARY(redisai_obj OBJECT
         execution/command_parser.c
         execution/parsing/deprecated.c
         execution/parsing/dag_parser.c
+        execution/parsing/tensor_commands_parsing.c
         execution/parsing/model_commands_parser.c
         execution/parsing/script_commands_parser.c
         execution/parsing/parse_utils.c

src/backends/libtorch_c/torch_c.cpp

@@ -198,7 +198,7 @@ void deleter(DLManagedTensor *arg) {
 }
 
 DLManagedTensor *toManagedDLPack(const torch::Tensor &src_) {
-    ATenDLMTensor *atDLMTensor(new ATenDLMTensor);
+    auto *atDLMTensor(new ATenDLMTensor);
     atDLMTensor->handle = src_;
     auto &src = atDLMTensor->handle;
     atDLMTensor->tensor.manager_ctx = atDLMTensor;
@@ -213,6 +213,7 @@ DLManagedTensor *toManagedDLPack(const torch::Tensor &src_) {
     atDLMTensor->tensor.dl_tensor.dtype = getDLDataType(src);
     atDLMTensor->tensor.dl_tensor.shape = const_cast<int64_t *>(src.sizes().data());
     atDLMTensor->tensor.dl_tensor.strides = const_cast<int64_t *>(src.strides().data());
+    atDLMTensor->tensor.dl_tensor.elements_length = nullptr;
     atDLMTensor->tensor.dl_tensor.byte_offset = 0;
     return &(atDLMTensor->tensor);
 }