Upstream: ggml-org#28656
Two identical multimodal inputs in one /embeddings request return different vectors under the server's default unified KV cache.
Environment: Linux aarch64, NVIDIA GB10 / CUDA 13, current main 434ddbbc0e30522e897670681e503b797c12b7c1; ggml-org/tinygemma3-GGUF Q8_0 plus its mmproj and the repo's tools/mtmd/tests/test-1-positive.png.
LLAMA_MEDIA_MARKER='<__media__>' llama-server \
-m tinygemma3-Q8_0.gguf --mmproj mmproj-tinygemma3.gguf \
-ngl 99 --embeddings -c 4096 -b 512 --seed 42 --no-jinja
Send one POST /embeddings whose content contains the same item twice:
{"prompt_string":"What is this: <__media__>\n","multimodal_data":["<same base64 PNG>"]}Observed: default kv_unified=true gives max|delta|=0.0705835, cosine 0.99989352. Adding -np 4 selects kv_unified=false and both vectors are bit-identical.
This isolates the failure to CUDA multimodal embeddings with the unified KV path; text-only and CPU are reported unaffected. Introducing commit not identified.
Upstream: ggml-org#28656
Two identical multimodal inputs in one
/embeddingsrequest return different vectors under the server's default unified KV cache.Environment: Linux aarch64, NVIDIA GB10 / CUDA 13, current main
434ddbbc0e30522e897670681e503b797c12b7c1;ggml-org/tinygemma3-GGUFQ8_0 plus its mmproj and the repo'stools/mtmd/tests/test-1-positive.png.LLAMA_MEDIA_MARKER='<__media__>' llama-server \ -m tinygemma3-Q8_0.gguf --mmproj mmproj-tinygemma3.gguf \ -ngl 99 --embeddings -c 4096 -b 512 --seed 42 --no-jinjaSend one
POST /embeddingswhosecontentcontains the same item twice:{"prompt_string":"What is this: <__media__>\n","multimodal_data":["<same base64 PNG>"]}Observed: default
kv_unified=truegivesmax|delta|=0.0705835, cosine0.99989352. Adding-np 4selectskv_unified=falseand both vectors are bit-identical.This isolates the failure to CUDA multimodal embeddings with the unified KV path; text-only and CPU are reported unaffected. Introducing commit not identified.