Cormac Shannon
84b04d12e9
Pipeline works on MPS (~105 s per 300-step run), results are reproducible, and the divergence stop works. Not experimental evidence. Claude-Session: https://claude.ai/code/session_01S8KN7NuUCm8nXLmy6eDuVv
7 lines
983 B
CSV
7 lines
983 B
CSV
run_id,status,qk_norm,learning_rate,n_layer,n_embd,seed,max_iters,params_M,best_val_loss,diverged_at,wall_s,git_commit,dirty,note
|
|
20261028-180656_spike-qkoff-lr1e-3,completed,False,0.001,4,256,1337,300,16.03,3.459,,102.9,55ba128dcc,False,tech spike smoke test (not an experiment)
|
|
20261028-180840_spike-qkon-lr1e-3,completed,True,0.001,4,256,1337,300,16.03,3.4304,,104.8,55ba128dcc,False,tech spike smoke test (not an experiment)
|
|
20261028-181025_spike-qkoff-lr1e-1,completed,False,0.1,4,256,1337,300,16.03,5.2948,,103.4,55ba128dcc,False,tech spike smoke test (not an experiment); divergence-detection check
|
|
20261028-181209_spike-qkon-lr1e-1,completed,True,0.1,4,256,1337,300,16.03,3.9026,,105.7,55ba128dcc,False,tech spike smoke test (not an experiment); divergence-detection check
|
|
20261028-181408_test-divergence-stop,diverged,False,0.1,4,256,1337,60,16.03,10.9167,0,5.8,55ba128dcc,False,plumbing test of the divergence stop (threshold set artificially low); not an experiment
|