graph_nemi_rl_e3_best: pure RL stage 3 from graph_nemi_sft_e2, best eval pass@1 3.04% at step 280 (run paused at step 300) 9233b00 verified JamesK2W commited on about 1 hour ago
her_nemi_graph_sft_e3_ck174: relabel study: main mix data, batch 64, epoch 2 of 3; eval pass@8 11.1% (20.9/13.4/2.8), pass@1 1.99% edcea99 verified JamesK2W commited on about 2 hours ago
her_nemi_graph_343_sft_e3_ck300: relabel study: 3:4:3 bands, 1024 reach + 512 near, batch 64, epoch 3 of 3; eval pass@8 12.0% (21.6/12.0/6.0), pass@1 2.17% 198f6fc verified JamesK2W commited on about 2 hours ago
her_nemi_graph_sft_e3: relabel study: 1024 reach + 512 near (scene graph), from graph_nemi_sft_e2, 510 steps batch16; eval pass@8 12.0% (21.6/13.9/4.1), pass@1 1.92% d38a192 verified JamesK2W commited on about 2 hours ago
smoke_p2v_rl_final: final checkpoint (global_step_2 of 2 steps) 1fbc95b verified JamesK2W commited on 1 day ago
graph_nemi_sft_e2: SFT from scene_nemi_rl_e2_final on graph-stitched near-miss demos of the RL e2 rollouts; eval pass@8 11.1% 5a35a84 verified JamesK2W commited on 1 day ago
rename graph_submitonly_epoch_0 -> graph_submitonly_sft_e0 701e45c verified JamesK2W commited on 1 day ago