AX EmbeddingGemma 300M MLX 8-bit
Parameter count: approximately 302.86M logical parameters.
8-bitis the quantization precision, not the model size.
Modified-distribution notice: AutomatosX added this model card, model-manifest.json, ax_provenance.json, LICENSE, and NOTICE. The MLX model tensor payload, configuration, tokenizer, and Sentence Transformers assets are unchanged from the upstream conversion. Safetensors header metadata was added only to report the logical parameter count to the Hub.
This is an MLX text-embedding model for Apple Silicon. The weight tensor payload and the other upstream runtime assets are byte-identical to mlx-community/embeddinggemma-300m-8bit at revision c8316c9e35cf8830541b04d315398c67faa6e497.
AutomatosX adds an AX Engine native manifest, pinned provenance, and tested serving instructions. Tensor payload bytes were not changed or re-quantized. This is an embedding-only release and does not use MTP.
Model details
- Base conversion source: google/embeddinggemma-300m-qat-q8_0-unquantized
- Format: MLX Safetensors
- Quantization: 8-bit, group size 64
- Embedding dimension: 768
- Transformer layers: 24
- Configured context limit: 2,048 tokens
- Intended hardware: Apple Silicon
Download
This is a public gated model. Sign in to Hugging Face and acknowledge the Gemma terms on this repository page before downloading:
hf download AutomatosX/AX-EmbeddingGemma-300M-MLX-8bit \
--local-dir ./AX-EmbeddingGemma-300M-MLX-8bit
Serve with AX Engine
Install AX Engine, then serve the downloaded directory:
ax-engine serve ./AX-EmbeddingGemma-300M-MLX-8bit \
--port 31418 -- --model-id embeddinggemma
Use the task prefixes expected by EmbeddingGemma. For retrieval, use the query prefix for queries and the document prefix for corpus passages:
curl http://127.0.0.1:31418/v1/embeddings \
-H 'Content-Type: application/json' \
--data '{
"model": "embeddinggemma",
"input": [
"task: search result | query: What is the capital of Canada?",
"title: none | text: Ottawa is the capital city of Canada."
],
"encoding_format": "float",
"pooling": "mean",
"normalize": true
}'
AX Engine uses EmbeddingGemma's bidirectional encoder, masked mean pooling, two-layer Dense projection head, and L2 normalization. The endpoint returns 768-dimensional vectors. Do not pass the OpenAI dimensions field to this AX Engine version.
Other useful prefixes are recorded in config_sentence_transformers.json, including sentence similarity, clustering, classification, retrieval, and summarization.
Validation and provenance
The release was validated on macOS arm64 with AX Engine 6.9.0:
- AX native artifact validation: ready, with no issues
- Live POST /v1/embeddings batch: passed
- Returned dimensions: 768
- L2-normalized vector norms: approximately 1.0
- Upstream tensor payload, configuration, tokenizer, and Sentence Transformers assets: byte-exact; the Safetensors header and shard index add only the corrected logical parameter count
See ax_provenance.json for pinned source and SHA-256 values.
License and use restrictions
This model is distributed under the Gemma Terms of Use, including the incorporated Gemma Prohibited Use Policy. LICENSE contains a copy of the agreement, and NOTICE contains the redistribution notice required by that agreement. Recipients must comply with those terms.