lsteno/Qwen3-4B-Instruct-2507-RLM-SFT-v3-per-root-turn

4B • Updated May 17 • 3

Note SFT attempt, per-root-turn trajectory format

lsteno/Qwen3-4B-Instruct-2507-RLM-RL-depth1-r4-a8-lr5e-7-s150-lora

Updated May 17 • 9

Note LoRA RL, depth-1, rank 4 alpha 8, LR 5e-7, 150 steps

lsteno/Qwen3-4B-Instruct-2507-RLM-RL-depth1-r4-a8-lr1e-5-s150-lora

Updated May 18 • 3

Note LoRA RL, depth-1, rank 4 alpha 8, LR 1e-5, 150 steps

lsteno/qwen3-rlm-depth1-r4-a8-lr1e-4-s150-bal35f40v1-lora

Updated May 19 • 4

Note LoRA RL, balanced BEEG, rank 4 alpha 8, LR 1e-4, 150 steps

lsteno/qwen3-rlm-depth1-r16-a32-lr5e-7-s150-bal35f40v1-lora

Updated May 19 • 3

Note LoRA RL, balanced BEEG, rank 16 alpha 32, LR 5e-7, 150 steps

lsteno/qwen3-rlm-depth1-r16-a32-lr1e-5-s150-bal35f40v1-lora

Updated May 20 • 2

Note LoRA RL, balanced BEEG, rank 16 alpha 32, LR 1e-5, 150 steps

lsteno/qwen3-rlm-depth1-r16-a32-lr1e-4-s150-bal35f40v1-lora

Updated May 20 • 2

Note LoRA RL, balanced BEEG, rank 16 alpha 32, LR 1e-4, 150 steps

lsteno/qwen3-rlm-depth1-r64-a128-lr5e-7-s150-bal35f40v1-lora

Updated May 20 • 8

Note LoRA RL, balanced BEEG, rank 64 alpha 128, LR 5e-7, 150 steps

lsteno/qwen3-rlm-depth1-r64-a128-lr1e-5-s150-bal35f40v1-lora

Updated May 21 • 2

Note LoRA RL, balanced BEEG, rank 64 alpha 128, LR 1e-5, 150 steps

lsteno/Qwen3-4B-Instruct-2507-RLM-RLVR-FullFT-lr5e-6-depth1-v1

Text Generation • 4B • Updated May 23 • 18

Note Full fine-tune RLVR, depth-1, LR 5e-6, 150 steps

lsteno/Qwen3-4B-Instruct-2507-RLM-RLVR-FullFT-lr1e-5-depth1-v1

4B • Updated May 23 • 3

Note Full fine-tune RLVR, depth-1, LR 1e-5, 150 steps

Note BEEG-agents dataset used for RLM RLVR train/eval splits

lsteno/Qwen3-4B-Instruct-2507-RLM-RLVR-depth2-recursive-r64-a128-lr1e-5-adapter

Reinforcement Learning • Updated Jun 13 • 3

Note Depth-2 recursive RLM RLVR LoRA adapter, r64/a128/lr1e-5, step 150.