← Merchant ticket triage (BG/EN), guidelines v2 + real tickets (ABCD train)
Playground
Send one input to the base model and to the tuned adapter and compare the answers.
On this demo the playground runs on CPU: the base model and run 6 iteration 2 as GGUF q8_0 builds (llama.cpp), not the bf16 models the GPU evaluated. Their own scores are in docs/demo-run-log.md (R4). Runs without a served adapter show the base model only.
Hold-out examples (never used for training or evaluation)
Asking both models...
Base model
No prediction yet.
Tuned adapter
No prediction yet.