Commit History

card: add 'Final state, all tiers' summary
d894a9c
verified

avlp12 commited on

card: add tier-by-tier tuning rationale (why only T128 got sensitivity-placement + ZH-DWQ)
c713d75
verified

avlp12 commited on

T128 v2: sensitivity-ranked down3 + ZH-mixed DWQ (KL 0.802->0.550, ZH 1.43->0.83), tag T128-v1
784730b
verified

avlp12 commited on

fig3: T128 v2 numbers
4777461
verified

avlp12 commited on

MTP notes: corrected 4-bit draft analysis (measured draft latency + per-prompt acceptance)
6e9a8d4
verified

avlp12 commited on

MTP section: measured self-spec results (k=1 1.13x lossless) + harness usage
e9c7d76
verified

avlp12 commited on

T128 branch now DWQ-distilled: KL 0.942->0.802, all tables/figures updated
bb2a2c7
verified

avlp12 commited on

fig3: T128 DWQ-distilled values
2cf9104
verified

avlp12 commited on

model card: head-to-head comparison vs 4 other Hy3 MLX ports (same harness)
9f52141
verified

avlp12 commited on

model card: add figures (tier fit, recipe, KL-by-language)
b48c24e
verified

avlp12 commited on

card figures
2a59fd7
verified

avlp12 commited on

card figures
fd84969
verified

avlp12 commited on

card figures
4a880fe
verified

avlp12 commited on

model card: 4-tier table, measured quality/needle/memory, usage, T128 experimental warning
a07beed
verified

avlp12 commited on

Add files using upload-large-folder tool
d0b77a0
verified

avlp12 commited on

initial commit
caa267a
verified

avlp12 commited on