-
add the crossed Qwen3.8, Nemotron-3.5-Lightning and Inkling-Small runs (training_runs, splits)
-
add the Inkling-Small runs (training_runs, splits)
-
add the Qwen3.8 and Nemotron-3.5-Lightning runs (training_runs, splits, weights)
-
default = the post's models; other DeepSeek/Nemotron runs in default_other_models
-
Kimi/Qwen3.8 runs in their own config (viewer: max 30 splits per config); Qwen3.8 high-risk rows
-
All exp04 temptation evals: configs by prompt set x backend, one split per model; judge_input_truncated
-
data + card
-
data + card