Issues

jax_ile: --angle-marg-scheme multipeak (#313) is not traceable under the sampler jit; Table 3 arms cannot run it

#316

· oshaughnessy-junior opened

on Sep 9, 2026

four-axis policy: local branch accepts +16 nat at rho 163 under the #301 capacities (max_time_nodes 256, max_starts 128, budget 1e-2)

#308

· oshaughnessy-junior opened

on Sep 9, 2026

jax gate: test_jax_slowrot segfaults on CIT when test_flow_reuse_default.py is collected first (pre-existing on rift_O4d)

#300

· oshaughnessy-junior opened

on Sep 9, 2026

jax gate: CI runs an unpinned jax that no local environment has

#292

· oshaughnessy-junior opened

on Sep 8, 2026

anglemarg buffer cap does not cover the flowMC sampling path: 833 GiB request, laplace unusable under flowmc-phipsimarg

#253

· oshaughnessy-junior opened

on Sep 5, 2026

_runtime_amp_failsafe's SUSPECT label is uninformative under vmap (i.e. on every production path)

#240

· oshaughnessy-junior opened

on Sep 3, 2026

Vectorized non-GPU time marginalization offsets by ONE detector's max while exponentiating the accumulated network sum (platform-conditional; float64 platforms get #232's 745-nat budget)

#236

· oshaughnessy-junior opened

on Sep 3, 2026

batchmode defaults --interpolate-time to 'nearest', which its own help text says is never competitive — and the jax driver defaults to 'sinc'

#233

· oshaughnessy-junior opened

on Sep 3, 2026

Vectorized time-marginalized likelihood offsets by the BATCH max, not the per-sample max: lnL underflows to -inf above rho ~ 40 and mcsamplerAV collapses (device-independent; affects a delivered O4b result)

#232

· oshaughnessy-junior opened

on Sep 3, 2026

AV + --distance-marginalization without --force-xpy: TypeError on every point, and the driver exits 0 with no output files

#231

· oshaughnessy-junior opened

on Sep 3, 2026

mcsamplerGPU reports a mismatched CUDA stack as "no cupy" and silently runs numpy under --gpu --force-xpy (162 of 171 runs in one campaign)

#229

· oshaughnessy-junior opened

on Sep 2, 2026

test/jax: seven slowrot_fs_lib callers silently inherit a 4x finer rholm sampling

#197

· oshaughnessy-junior opened

on Aug 26, 2026

Navigation Menu

All issues

jax_ile: --angle-marg-scheme multipeak (#313) is not traceable under the sampler jit; Table 3 arms cannot run it

four-axis policy: local branch accepts +16 nat at rho 163 under the #301 capacities (max_time_nodes 256, max_starts 128, budget 1e-2)

jax gate: test_jax_slowrot segfaults on CIT when test_flow_reuse_default.py is collected first (pre-existing on rift_O4d)

jax gate: CI runs an unpinned jax that no local environment has

anglemarg buffer cap does not cover the flowMC sampling path: 833 GiB request, laplace unusable under flowmc-phipsimarg

_runtime_amp_failsafe's SUSPECT label is uninformative under vmap (i.e. on every production path)

Vectorized non-GPU time marginalization offsets by ONE detector's max while exponentiating the accumulated network sum (platform-conditional; float64 platforms get #232's 745-nat budget)

batchmode defaults --interpolate-time to 'nearest', which its own help text says is never competitive — and the jax driver defaults to 'sinc'

Vectorized time-marginalized likelihood offsets by the BATCH max, not the per-sample max: lnL underflows to -inf above rho ~ 40 and mcsamplerAV collapses (device-independent; affects a delivered O4b result)

AV + --distance-marginalization without --force-xpy: TypeError on every point, and the driver exits 0 with no output files

mcsamplerGPU reports a mismatched CUDA stack as "no cupy" and silently runs numpy under --gpu --force-xpy (162 of 171 runs in one campaign)

test/jax: seven slowrot_fs_lib callers silently inherit a 4x finer rholm sampling

Search results