Kept / trial active in llama-swap.yaml
| Run | Score | tok/step | R5 | Log |
|---|---|---|---|---|
| No results-*.log found for this id/alias. | ||||
"embed":
# raggo.net embeddings (raglib.py, model id "embed"). Qwen3-Embedding-0.6B,
# 1024-dim. CPU-only (-ngl 0) so it never competes with the resident GPU chat
# model for VRAM. --embedding + --pooling last (Qwen3-Embedding uses last-token
# pooling); the asymmetric query-instruction prefix is applied by raglib, not here.
# Kept resident by the embed-persistent group below (never swaps out).
cmd: |
${llama_bin}
-hf Qwen/Qwen3-Embedding-0.6B-GGUF:Q8_0
--alias embed
--embedding --pooling last
-ngl 0 -c 8192
--host 127.0.0.1 --port 9108 --parallel 2
proxy: http://127.0.0.1:9108