Kakuto AITry Now

Lab notebook

Model lineage

Every model here started from an earlier one. A skill is trained on top of the last good model; when several techniques need learning at once, the base is cloned into teachers, one technique each, and then the teachers are distilled back into one student. Each box below is a training run, with an arrow from the run it started from and a dashed one from each teacher it learnt from.

Each box is a stage of the work, a work log: a retrain, a set of teachers, a distillation. Open one for every run in it. Each technique's page has the same picture run by run, for its own runs.

The stages of the work, oldest on the left: 21 work logs, with an arrow where a model from one stage started a run in another, dashed where it was a teacher.FoundationsTechniquesStudentsRun by run: balancing in the stance: Every run behind the post, with its outcome, machines and cost.Balancing in the stance19 runs · $23 · 13 keptRun by run: the jab: Every run behind the post, with its outcome, machines and cost.Jab22 runs · $16 · 14 keptRun by run: five punches: Every run behind the post, with its outcome, machines and cost.Five punches10 runs · $22 · 7 keptRun by run: faster punches: Every run behind the post, with its outcome, machines and cost.Faster punches7 runs · $8 · 5 keptRun by run: walking in the stance: Every run behind the post, with its outcome, machines and cost.Walking in the stance22 runs · $35 · 18 keptRun by run: blocking: Every run behind the post, with its outcome, machines and cost.Blocking20 runs · $33 · 20 keptFast stepping and the step-jab (abandoned): Twenty-six short GPU runs on fast steps in and out of range and a jab that steps in. The long step never pushed off the back leg, and the work was stopped to redo the motion capture.Fast stepping and the ste…25 runs · $9 · 18 keptThe higher guard: retraining every skill onto a new stance: Eighteen runs that moved every skill onto a higher guard, fixed the lead hook and the jab, and produced guard18-124.5M, the base every clone starts from.The higher guard: retrain…18 runs · $21 · 8 keptTeacher: the rear straight to the body: Four runs to a body-straight teacher: a no-low-blow line, the bag at the impact point, then a retrain on the base's slower elbows.Rear straight to the body4 runs · $4 · 4 keptTeachers: the rear knee and the switch knee: Two knee teachers trained side by side, three runs each: the knees landed first time, the feet and the steps home took two more.Rear knee and the switch…6 runs · $6 · 6 keptDistillation: six runs to the student distill4: Distilling the body straight, both knees and the pull counter into the base: what each run changed, how the gate picked the student, and what it cost to evaluate.Six runs to the student d…6 runs · $16 · 3 keptTeacher: the pull counter: One run on RunPod after four vast.ai boxes never loaded: a lean back and a rear straight to the head, 8/8 in the air and at the bag.Pull counter1 runs · $3 · 1 keptTeachers: slips, rolls and the lean back: Nine evasion teachers in two rounds of four parallel boxes. They passed the gate and failed on video: rushed by the punch clock, then bent by an edited reference.Slips, rolls and the lean…9 runs · $13 · 1 keptTeachers: the uppercuts over distance: Two uppercut teachers trained in parallel on RunPod. The rear one was kept; the lead one turned 90° after the punch, which traced back to how training rewarded facing.Uppercuts over distance2 runs · $8 · 1 keptTeachers: the uppercuts, second round: Four more uppercut teachers: the rear one kept facing square, the lead one still turned or staggered until its feet were held down.Uppercuts, second round4 runs · $10 · 2 keptTeacher: the rear knee, landing upright: One more rear-knee teacher, so the knee lands upright instead of arched back.Rear knee, landing upright1 runs · $1 · 1 keptTeacher: the dash: Six runs to a dash teacher: a quick push-off in any direction, upright, without squatting, over the distances I would really cover.Dash6 runs · $10 · 4 keptTeachers: the kicks: Low, mid and high roundhouses on both sides, from my motion capture: the first teachers, and the next round prepared.Kicks6 runs · $14 · 4 kept · goingDistillation: distill7 and distill8: Two more students: the uppercuts over distance and the upright knee in distill7, then the elbows in distill8.Distill7 and distill82 runs · $6 · 1 kept · goingTeachers: the elbows: Two elbow teachers from my motion capture: the right elbow to the chin and the stepping left elbow.Elbows2 runs · $7 · 2 kept · goingThe muscle body: muscles instead of joint motors: An experiment I asked for: drive the fighter with 86 joint muscles instead of servos, and measure how much longer the stance takes to learn, if it can be learnt at all.The muscle body: muscles…3 runs · $0 · going
a model from it started runs hereit taught runs hereOldest on the left. Each box is a work log: open it for every run, or scroll sideways.