Eval bug: the vector of jina-embeddings-v2-base-zh retrieval result is incorrect.

[zhaojc001]

Name and Version

./build/bin/llama-cli --version
version: 7449 (6853bee)
built with GNU 11.4.0 for Linux x86_64

Operating systems

Linux

GGML backends

CPU

Hardware

AMD EPYC 9K84 96-Core Processor

Models

jina-embeddings-v2-base-zh

Problem description & steps to reproduce

1.Convert HF model to GGUF model：
/app/jczhao/miniconda3/envs/llamacpp/bin/python convert_hf_to_gguf.py
/app/jczhao/jina-embeddings-v2-base-zh
--outfile /app/jczhao/jina-embeddings-v2-base-zh-f32.gguf
--outtype f32
--metadata "general.architecture=encoder"
--metadata "encoder.context_length=8192"
--metadata "general.name=jina-embeddings-v2-base-zh"
--verbose
2.Llama server deployment command：
sudo nohup ./build/bin/llama-server -m /app/jczhao/jina-embeddings-v2-base-zh-f32.gguf -t 24 -c 8192 -b 512 --host 0.0.0.0 --port 7868 --embeddings --cont-batching --no-mmap > llama-server-7.log 2>&1 &
Compared to the official HF model and ONNX model,the vector of jina-embeddings-v2-base-zh gguf model retrieval result is incorrect!
Please help me solve the issue,thanks!

First Bad Commit

No response

Relevant log output

Logs