ept-lm-page.git, branch main Unnamed repository; edit this file 'description' to name the repository. Replace the EP-family wall-clock figure with the measured backprop ratio 2026-08-03T14:19:01+00:00 YurenHao0426 Blackhao0426@gmail.com 2026-08-03T14:19:01+00:00 4e32a4f9ea69f9b814fa0363a31ecf9224294477 The 0.4x figure compared relaxation-iteration counts across methods that count different things: the closest EP-family baseline uses asynchronous local updates where one iteration moves information roughly one layer, while ours transmits through the whole stack per round. Wall-clock against backprop on matched hardware is the currency that survives that difference, and it measures 2.7x. 
The 0.4x figure compared relaxation-iteration counts across methods that count
different things: the closest EP-family baseline uses asynchronous local updates
where one iteration moves information roughly one layer, while ours transmits
through the whole stack per round. Wall-clock against backprop on matched
hardware is the currency that survives that difference, and it measures 2.7x.
Update results to the fixed-readout numbers and add the precision-floor finding 2026-08-02T18:00:27+00:00 YurenHao0426 Blackhao0426@gmail.com 2026-08-02T18:00:27+00:00 6943da547fce08acc70647d1a93988b8585eedcd 72M is now 3.400 vs 3.397, statistically indistinguishable rather than a 4 to 5 percent band, with the 11M, 18M and 36M rungs reported alongside. Adds the readout precision result and its verification, and marks the 135M rerun as in progress. 
72M is now 3.400 vs 3.397, statistically indistinguishable rather than a 4 to 5 percent band, with the 11M, 18M and 36M rungs reported alongside. Adds the readout precision result and its verification, and marks the 135M rerun as in progress.
Make the EP explanation intuitive (#7) 2026-07-30T08:05:21+00:00 Yuren Hao 97327730+YurenHao0426@users.noreply.github.com 2026-07-30T08:05:21+00:00 c29d686be44e4660c314eb1883f80cb0a0023224 






Trim the result label (#6)
2026-07-30T08:02:47+00:00

Yuren Hao
97327730+YurenHao0426@users.noreply.github.com

2026-07-30T08:02:47+00:00

bc813c1e5936f90f4cef7b07d4bc86eb7ca3645d












Strengthen the BP-match and scaling-roadmap framing (#5)
2026-07-30T08:00:28+00:00

Yuren Hao
97327730+YurenHao0426@users.noreply.github.com

2026-07-30T08:00:28+00:00

66c6966cd7209d84e9ace9ccfaa8d907a2e3c6f2

* Frame the 72M result as a BP match

* Clarify the 0.6B to 8B scaling roadmap




* Frame the 72M result as a BP match

* Clarify the 0.6B to 8B scaling roadmap






Frame training cost against EP baselines (#4)
2026-07-30T07:58:33+00:00

Yuren Hao
97327730+YurenHao0426@users.noreply.github.com

2026-07-30T07:58:33+00:00

220cc883ecb2f007edc7ecd90fd51d2a2b7be208












Restore project team credits (#3)
2026-07-30T07:55:00+00:00

Yuren Hao
97327730+YurenHao0426@users.noreply.github.com

2026-07-30T07:55:00+00:00

1f62b37b2b93e3563bab970d0f280889e38849d2












Add project favicon (#2)
2026-07-30T07:53:25+00:00

Yuren Hao
97327730+YurenHao0426@users.noreply.github.com

2026-07-30T07:53:25+00:00

c888e54d7e1f5d8d30406ea5b0b6bafb4cdf9f22












Rewrite grant project page (#1)
2026-07-30T07:51:50+00:00

Yuren Hao
97327730+YurenHao0426@users.noreply.github.com

2026-07-30T07:51:50+00:00

14e2adbd11c75098d74f1733f28bbeb9443891a7












Abstract: same doctrine
2026-07-30T01:43:00+00:00

YurenHao0426
Blackhao0426@gmail.com

2026-07-30T01:43:00+00:00

2ac145ed97ea81d0606c7c5b9fa71fd3e7336296

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn





Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn