F2LLM Embedding v2 80M. Qwen3-based, pruned from the 0.6B base; the smallest member of the family.
Cosine similarity vs the upstream sentence-transformers reference on a fixed
test set (text):
1# Download
2huggingface-cli download cstr/f2llm-v2-80m-GGUF f2llm-v2-80m-q8_0.gguf --local-dir .
3
4# Run with CrispEmbed
5./crispembed -m f2llm-v2-80m-q8_0.gguf "Hello world"
6
7# Or with auto-download
8./crispembed -m f2llm-v2-80m "Hello world"
Verified bit-identical to HuggingFace sentence-transformers (cosine similarity >= 0.999 on test texts).
CrispEmbed is a lightweight C/C++ text embedding inference engine using ggml.
No Python runtime, no ONNX. Supports BERT, XLM-R, Qwen3, and Gemma3 architectures.
1# Build CrispEmbed
2git clone https://github.com/CrispStrobe/CrispEmbed
3cd CrispEmbed
4cmake -S . -B build && cmake --build build -j
5
6# Encode
7./build/crispembed -m f2llm-v2-80m-q8_0.gguf "query text"
8
9# Server mode
10./build/crispembed-server -m f2llm-v2-80m-q8_0.gguf --port 8080
11curl -X POST http://localhost:8080/v1/embeddings \
12 -d '{"input": ["Hello world"], "model": "f2llm-v2-80m"}'
