notices
benchmarks share of test messages with every field exactly right · none of these messages was trained on
training
trainer log support-slm-v2
feed
training rows
trainer log support-slm-v1
feed
evaluation · v1 (GGUF) exact match, every field
prediction feed support-slm v1 (GGUF), sampled from the three test sets, raw model output
feed
per-field accuracy · held-out 420 messages