- Sentinel-anchored unconditional pass: both branches now predict the same N response tokens - Single-token responses (rl=1) fully supported - ctx_len tracked per sample; skip_reason replaces silent None - --per_token flag for per-token IFD breakdown |
||
|---|---|---|
| .. | ||
| demo | ||
| eval | ||
| tools | ||
| docker.sh | ||
| pre_commit.sh | ||