failed when changing GPU delegate compatibility database.

[gaikwadrahul8]

1. System information

OS Platform and Distribution (e.g., Linux Ubuntu 18.04):
TensorFlow Source Code 2.9

2. Code
   When following /home/ts/tensorflow_src/tensorflow/lite/experimental/acceleration/compatibility/README.md to change delegate compatibility database,

When convertting from json to flatbuffer flatc -b database.fbs -- gpu_compatibility.json, it gave that
Usage: flatc [-b|--binary, -c|--cpp, -n|--csharp, -d|--dart, -g|--go, -j|--java,
-t|--json, --jsonschema, --kotlin, --lobster, -l|--lua, --nim, --php, --proto,
-p|--python, -r|--rust, --swift, -T|--ts, -o, -I, -M, --version, -h|--help,
--strict-json, --allow-non-utf8, --natural-utf8, --defaults-json,
--unknown-json, --no-prefix, --scoped-enums, --no-emit-min-max-enum-values,
--swift-implementation-only, --gen-includes, --no-includes, --gen-mutable,
--gen-onefile, --gen-name-strings, --gen-object-api, --gen-compare,
--gen-nullable, --java-package-prefix, --java-checkerframework, --gen-generated,
--gen-jvmstatic, --gen-all, --gen-json-emit, --cpp-include, --cpp-ptr-type,
--cpp-str-type, --cpp-str-flex-ctor, --cpp-field-case-style, --cpp-std,
--cpp-static-reflection, --object-prefix, --object-suffix, --go-namespace,
--go-import, --go-module-name, --raw-binary, --size-prefixed,
--proto-namespace-suffix, --oneof-union, --keep-proto-id, --proto-id-gap,
--grpc, --schema, --bfbs-filenames, --bfbs-comments, --bfbs-builtins,
--bfbs-gen-embed, --conform, --conform-includes, --filename-suffix,
--filename-ext, --include-prefix, --keep-prefix, --reflect-types,
--reflect-names, --rust-serialize, --rust-module-root-file, --root-type,
--require-explicit-ids, --force-defaults, --force-empty, --force-empty-vectors,
--flexbuffers, --no-warnings, --warnings-as-errors, --cs-global-alias,
--cs-gen-json-serializer, --json-nested-bytes, --ts-flat-files,
--ts-entry-points, --annotate-sparse-vectors, --annotate,
--no-leak-private-annotation]... FILE... [-- BINARY_FILE...]

error:
current schema has no file_identifier: cannot test if "gpu_compatibility.json" matches the schema, use --raw-binary to read this file anyway.

3. and then using flatc -b database.fbs -- gpu_compatibility.json --raw-binary to convert.
4. Found that convertted gpu_compatibility.bin size is 117K, but origin one is only 17K.
5. and then compile
   bazel build -c opt --config=android_arm64 //tensorflow/lite/java:tensorflowlite_gpu
   bazel build -c opt --config=android_arm64 //tensorflow/lite/java:libtensorflowlite_gpu_jni.so

Using libtensorflowlite_gpu_jni.so in example image classification android app, and the app was crashed.

Very thanks.

[gaikwadrahul8]

This issue originally reported by @megleo has been moved to this dedicated repository for LiteRT to enhance issue tracking and prioritization. To ensure continuity, we have created this new issue on your behalf.

We appreciate your understanding and look forward to your continued involvement.

[pkgoogle]

Original Issue: tensorflow/tensorflow#60079