all-MiniLM-L6-v2 (MNN, GGUF)

The sentence-transformers all-MiniLM-L6-v2 embedding model in MNN form, which TokForge keeps as its fallback embedder for document and memory search.

Published by TokForge for on-device inference.

TokForge

• Website: https://tokforge.ai
• Discord: https://discord.gg/Acv3CBtfVm
• Google Play: https://play.google.com/store/apps/details?id=dev.tokforge
• iOS TestFlight: https://testflight.apple.com/join/jnufjzRr

Runs on-device in the TokForge app.

Use it in TokForge

1. Install TokForge from Google Play.
2. On Android, open Settings > Speed & engine > Models & Modules and download MiniLM L6 v2 Embedder.
3. You rarely need it by hand. TokForge's main embedder is bge-small-en-MNN. TokForge uses MiniLM as a fallback when BGE-small is not on the phone yet, and to keep searching documents and memories that an older version of the app embedded with MiniLM. It fetches minilm.mnn and vocab.txt on its own when it needs them and you allow it.

Collections: TokForge Voice and Helper Models.

Files

TokForge downloads only minilm.mnn and vocab.txt. The GGUF files are for llama.cpp, and model.onnx and vocab.txt are unchanged copies of the files in sentence-transformers/all-MiniLM-L6-v2.

About the model

all-MiniLM-L6-v2 maps sentences and short paragraphs to 384-dimensional vectors for semantic search, clustering and sentence similarity. It is a 6-layer MiniLM (BERT-style) encoder, trained by sentence-transformers from Microsoft's MiniLM-L6-H384-uncased. TokForge uses mean pooling, L2-normalized vectors and up to 128 tokens per input.

How it was made

minilm.mnn was converted with MNN's converter, and the GGUF files were made with llama.cpp.

Limitations

• English only
• Superseded by BGE-small in TokForge; kept for older installs and for vectors stored by older versions
• TokForge reads up to 128 tokens per input (BGE-small reads up to 512)

License

all-MiniLM-L6-v2 is released by sentence-transformers under the Apache-2.0 license. The MiniLM-L6-H384-uncased model it was trained from is released by Microsoft under the MIT license.

Credits

• all-MiniLM-L6-v2 by sentence-transformers (Hugging Face). Used under the Apache-2.0 license.
• MiniLM-L6-H384-uncased by Microsoft (nreimers) (Hugging Face). Used under the MIT license.
• Converted to MNN with MNN by Alibaba (Apache-2.0); GGUF files made with llama.cpp (MIT).