| # EmbeddingGemma 300M LiteRT-LM Configuration | |
| # Use this with the litertlm_builder_cli to create .litertlm file | |
| # | |
| # Usage: | |
| # cd deps/LiteRT-LM | |
| # bazel run //schema/py:litertlm_builder_cli -- \ | |
| # toml --path ../../models/embeddinggemma-300m.toml \ | |
| # output --path ../../models/embeddinggemma-300m.litertlm | |
| [system_metadata] | |
| entries = [ | |
| { key = "model_name", value_type = "String", value = "EmbeddingGemma-300M" }, | |
| { key = "model_version", value_type = "String", value = "1.0" }, | |
| { key = "embedding_dim", value_type = "Int32", value = 256 }, | |
| { key = "author", value_type = "String", value = "Google" } | |
| ] | |
| # Section 1: TFLite Embedder Model | |
| [[section]] | |
| section_type = "TFLiteModel" | |
| model_type = "EMBEDDER" | |
| # Use the seq512 version (best for tool descriptions) | |
| data_path = "/home/mac/git/mcp-agent/models/embeddinggemma-300M_seq512_mixed-precision.tflite" | |
| additional_metadata = [ | |
| { key = "embedding_dimensions", value_type = "Int32", value = 256 }, | |
| { key = "max_seq_length", value_type = "Int32", value = 512 } | |
| ] | |
| # Section 2: HuggingFace Tokenizer (if available) | |
| # Uncomment if you have the tokenizer.json | |
| # [[section]] | |
| # section_type = "HF_Tokenizer" | |
| # data_path = "tokenizer.json" | |