This website requires JavaScript.
9de0bad3d4
fix transformer: GQA text overflow, heatmap sizing, auto-regressive pos labels
main
ViperEkura
2026-05-25 19:19:48 +0800
d471cfa276
add hook.py: terminal demo + claim + logo reveal opening scene (~7s)
ViperEkura
2026-05-19 16:08:12 +0800
12d587aa92
refactor continuous_batching: 4-lane pipeline (PENDING/PREFILL/DECODE/FINISHED), remove bogus Trans arrow, Refill=admission per AstrAI arch
ViperEkura
2026-05-18 15:29:40 +0800
4f14d09fe3
fix paged_cache: _TaskRow class, expand rendering bug, and order corrected
ViperEkura
2026-05-09 20:16:44 +0800
bb0c32b032
redesign continuous batching: simplify to lane-based FSM with Prefill state
ViperEkura
2026-05-09 18:51:08 +0800
c9f290c3c8
Merge PROMO_GUIDE into README, add AGENTS.md, parallelize render_all
ViperEkura
2026-05-09 14:35:38 +0800
f5499866ad
refactor: replace prefix cache with paged KV cache across promo guide, architecture, and animation
ViperEkura
2026-05-08 22:38:14 +0800
740e766e99
refactor: remove summary text from prefix_cache, leave bottom space for subtitles
ViperEkura
2026-05-07 23:21:57 +0800
01ccf192ea
refactor: extract CTA scene into standalone cta.py
ViperEkura
2026-05-07 23:14:26 +0800
c2c18f10d6
fix: replace unicode checkmark with ASCII argmax label
ViperEkura
2026-05-07 23:08:16 +0800
e594d02d31
feat: merge D,E,F into one batch (gated by F at t=8) to show worse static idle
ViperEkura
2026-05-07 22:47:21 +0800
29007c7b97
chore: remove T0-T3 demo tokens from position-grouped decode section
ViperEkura
2026-05-07 22:25:25 +0800
6b0a1dbb5e
refactor: redesign batching FSM as queue pipeline with dynamic task states
ViperEkura
2026-05-07 17:56:17 +0800
c05a432e45
chore: use Times New Roman across all scenes, widen Transformer block to 2.4 for \'Transformer Block × 24\' label
ViperEkura
2026-05-07 14:59:27 +0800
57abefa47f
fix: shift GQA layout down 0.4 to avoid title-input overlap
ViperEkura
2026-05-07 14:51:31 +0800
6b26ec33ab
chore: switch .gitignore to whitelist mechanism (only .py, .md, .gitignore)
ViperEkura
2026-05-07 14:14:15 +0800
0018868ee3
refactor: transformer — heatmap two-phase scores+mask, auto-regressive full I/O pipeline with Emb, RMS Norm, LM Head, distribution
ViperEkura
2026-05-07 14:00:52 +0800
496f964979
refactor: architecture — boxes left, descriptions right, 4-layer layout
ViperEkura
2026-05-07 12:18:33 +0800
4d96a84fc5
refactor: architecture layer-by-layer introduction with source file refs
ViperEkura
2026-05-07 11:37:21 +0800
e7d736a3b0
fix: bottom spacing, remove specs card, full formula first with no ellipsis in steps
ViperEkura
2026-05-07 09:28:07 +0800
eeaf0a5a16
feat: SDPA formula breakdown + attention score heatmap with per-cell causal mask
ViperEkura
2026-05-07 00:58:46 +0800
ba100c19d1
refactor: Gantt chart — same 5 requests both sides, wait+run bars, batch boxes connect GPU↔reqs
ViperEkura
2026-05-07 00:33:42 +0800
b2ff243447
fix: call align_to AFTER move_to so bars follow rect positions
ViperEkura
2026-05-06 22:13:11 +0800
f6ccefb34e
fix: align throughput bars by same-x positioning instead of arrange Left-edge
ViperEkura
2026-05-06 22:11:18 +0800
ce2a7c165a
fix: throughput bars — proportional widths (1.0x vs 3.4x), proper GrowFromEdge anim
ViperEkura
2026-05-06 22:07:37 +0800
42c1d5f700
fix: throughput comparison layout — centered rows with proper Left-Right alignment
ViperEkura
2026-05-06 22:02:20 +0800
4ed16a70b4
fix: new requests enter at Prefill stage, not Refill
ViperEkura
2026-05-06 21:59:47 +0800
fc68fc9107
rewrite continuous_batching as state-machine with batch token rotation, in/out flow, bitmask anim, and throughput bars
ViperEkura
2026-05-06 21:53:15 +0800
a7a79eef96
fix: architecture layout and rewrite continuous_batching with dynamic pipeline animation
ViperEkura
2026-05-06 21:48:18 +0800
c03abd31fe
add project source files
ViperEkura
2026-05-06 21:16:57 +0800
94aadb3d8f
first commit
ViperEkura
2026-05-06 21:15:18 +0800