Ctrl+K
4B matched control (fixed-normal prior, identical SFT+RL recipe to hidden); folded LoRA, step 400
8c686c6 verified - qwen3_5_4b_base
- rl-matched-normal-step400
- rl-randprior-hidden-step400
- rl-randprior-hidden
- rl-randprior-told-step400
- rl-randprior-told
- rl-scalar-hidden-step400
- rl_9b_hidden_lr3e5_step90
- rl_lr1e5_step400
- sft-randprior-hidden
- sft-randprior-told
- sft_9b_matched_normal
- sft_9b_randprior_hidden
- sft_final
- verbalize-randprior-hidden-v1
- verbalize-randprior-told-v1
- verbalize_i4_v1
- 4.09 kB
- 1.32 kB