ViperEkura
/
video-promo
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Compare commits

base: ViperEkura:496f964979af5d39ce53e469fa23e3957a2d478e
Branches Tags
ViperEkura:main
..
compare: ViperEkura:e7d736a3b008809af4ffb35880367e74af3f667c
Branches Tags
ViperEkura:main

No commits in common. "496f964979af5d39ce53e469fa23e3957a2d478e" and "e7d736a3b008809af4ffb35880367e74af3f667c" have entirely different histories.
496f964979 ... e7d736a3b0

1 changed files with 52 additions and 111 deletions
Show Stats Expand all files Collapse all files

163
architecture.py
View File

@ -1,135 +1,76 @@
"""AstrAI promo: 4-layer architecture — boxes left, explanations right.""" """AstrAI promo: Full architecture overview."""
from manim import * from manim import *
class Architecture(Scene): class Architecture(Scene):
"""Boxes on left, description text on right for each layer.""" """Animates the full AstrAI system stack layer by layer."""
def construct(self): def construct(self):
title = Text("AstrAI Architecture", font_size=42, color=BLUE) title = Text("AstrAI Architecture", font_size=48, color=BLUE)
title.to_edge(UP, buff=0.25)
self.play(Write(title)) self.play(Write(title))
self.wait(0.2)
self.play(title.animate.to_edge(UP))
W, BH = 5.2, 1.15 layers_data = [
BX = -3.6 (0.9, GREEN, "API Layer", ["FastAPI Server • OpenAI-Compatible API"]),
TX = 3.6 (0.9, BLUE, "Inference Engine", ["Streaming • Async • Batch Modes"]),
(1.6, YELLOW, "Continuous Batching Scheduler",
def make_box(header, color, bits, src): ["Cleanup → Refill → Prefill → Decode",
b = Rectangle(width=W, height=BH, color=color, fill_opacity=0.1, stroke_width=1.5) "Position-Grouped Decode",
h = Text(header, font_size=16, color=color, weight=BOLD) "Bitmask O(1) Slot Allocation"]),
items = [h] (1.2, ORANGE, "Prefix Cache + KV Cache",
for line in bits: ["Radix Tree • Slot Versioning",
items.append(Text(line, font_size=10, color=WHITE)) "GPU copy_() → Zero-Copy Reuse"]),
items.append(Text(src, font_size=9, color=GRAY)) (1.2, PURPLE, "Transformer Model (1B params)",
c = VGroup(*items).arrange(DOWN, buff=0.04) ["24-layer GQA • RoPE • SwiGLU",
c.move_to(b.get_center()) "bfloat16 • 100K vocab"]),
return VGroup(b, c)
L1 = make_box("HTTP API Server", GREEN,
["FastAPI · OpenAI-Compatible",
"/v1/chat/completions · SSE streaming"],
"astrai/inference/server.py")
L2 = make_box("Inference Engine", BLUE,
["generate() · batch mode · streaming",
"4-phase daemon: Cleanup → Refill → Prefill → Decode",
"Position-grouped decode · Bitmask O(1) slots"],
"astrai/inference/engine.py · scheduler.py")
L3 = make_box("Prefix Cache + KV Cache", ORANGE,
["Radix Tree prefix matching · LRU eviction",
"Slot versioning · GPU copy_() zero-copy reuse"],
"astrai/inference/scheduler.py")
L4 = make_box("Transformer Model", PURPLE,
["24× DecoderBlock · GQA 6:1 · RoPE",
"SwiGLU MLP · Dim 1536 · bfloat16"],
"astrai/model/transformer.py")
layers = VGroup(L1, L2, L3, L4)
layers.arrange(DOWN, buff=0.08)
layers.move_to([BX, 0, 0])
layers.next_to(title, DOWN, buff=0.25)
# Description panels (right side)
descs_text = [
["HTTP API Server",
"Receives chat requests via",
"OpenAI-compatible endpoints.",
"Streams generated tokens back",
"through Server-Sent Events."],
["Inference Engine",
"Orchestrates the full generation",
"pipeline with a background daemon.",
"4-phase loop: Cleanup tasks,",
"Refill batch, Prefill prompts,",
"Decode tokens one by one."],
["Prefix Cache + KV Cache",
"Caches key-value states using",
"a Radix Tree for O(n) prefix lookup.",
"Reuses matched prefixes via GPU",
"memcpy — zero recomputation."],
["Transformer Model (1B params)",
"Decoder-only Transformer with",
"Grouped-Query Attention (GQA 6:1).",
"RoPE rotary encoding, SwiGLU",
"activation, 100K vocabulary."],
] ]
def make_desc(lines, color): layers = VGroup()
els = [Text(lines[0], font_size=20, color=color, weight=BOLD)] for height, color, label, subs in layers_data:
for ln in lines[1:]: box = Rectangle(width=7.5, height=height, color=color, fill_opacity=0.1)
els.append(Text(ln, font_size=14, color=WHITE)) lbl = Text(label, font_size=18, color=color)
grp = VGroup(*els).arrange(DOWN, buff=0.1, aligned_edge=LEFT) items = [lbl] + [Text(s, font_size=11, color=WHITE) for s in subs]
return grp content = VGroup(*items)
content.arrange(DOWN, buff=0.22)
content.move_to(box.get_center())
layers.add(VGroup(box, content))
COLORS = [GREEN, BLUE, ORANGE, PURPLE] layers.arrange(DOWN, buff=0.18)
descs = [make_desc(lns, c) for lns, c in zip(descs_text, COLORS)] layers.next_to(title, DOWN, buff=0.3)
arrows = VGroup() for i in range(len(layers)):
for i, (layer, desc) in enumerate(zip(layers, descs)): self.play(Create(layers[i]), run_time=0.35)
b = layer[0] if i > 0:
self.play(Create(layer), run_time=0.35) # Use box-to-box for arrow endpoints (not content)
desc.next_to(b, RIGHT, buff=1.0) prev_box = layers[i - 1][0]
desc.align_to(b, UP) curr_box = layers[i][0]
self.play(Write(desc), run_time=0.3)
self.wait(2.0 if i == 0 else 1.8)
if i < len(layers) - 1:
self.play(FadeOut(desc))
nxt = layers[i + 1][0]
arrow = Arrow( arrow = Arrow(
b.get_bottom(), nxt.get_top(), prev_box.get_bottom(),
color=GRAY, buff=0.04, curr_box.get_top(),
max_tip_length_to_length_ratio=0.18, color=GRAY,
buff=0.1,
max_tip_length_to_length_ratio=0.15,
) )
self.play(Create(arrow), run_time=0.12) self.play(Create(arrow), run_time=0.15)
arrows.add(arrow)
else:
self.wait(0.5)
self.play(FadeOut(desc))
# Show all boxes + arrows together briefly self.wait(0.5)
self.wait(0.3)
# Highlight innovation layers hl = SurroundingRectangle(layers[3], color=GREEN, buff=0.12)
hl2 = SurroundingRectangle(L2, color=BLUE, buff=0.1, stroke_width=2) hl_note = Text("Zero-Copy Prefix Reuse", font_size=18, color=GREEN)
hl3 = SurroundingRectangle(L3, color=ORANGE, buff=0.1, stroke_width=2) hl_note.next_to(hl, LEFT, buff=0.4)
hl_note = Text("Key Innovations", font_size=20, color=GOLD) self.play(Create(hl), Write(hl_note))
hl_note.next_to(VGroup(hl2, hl3), RIGHT, buff=1.5) self.wait(1.5)
hl_note.align_to(hl2, UP) self.play(FadeOut(hl), FadeOut(hl_note))
self.play(Create(hl2), Create(hl3), Write(hl_note))
self.wait(2.0)
self.play(FadeOut(hl2), FadeOut(hl3), FadeOut(hl_note))
self.play(FadeOut(VGroup(*layers)), FadeOut(arrows)) self.play(FadeOut(layers))
cta = VGroup( cta = VGroup(
Text("AstrAI", font_size=52, color=BLUE), Text("AstrAI", font_size=52, color=BLUE),
Text("Single GPU · Open Source · 1B params", font_size=22, color=GRAY), Text("Single GPU • Open Source • 1B params", font_size=24, color=GRAY),
Text("github.com/ViperEkura/AstrAI", font_size=18, color=YELLOW), Text("github.com/ViperEkura/AstrAI", font_size=20, color=YELLOW),
).arrange(DOWN, buff=0.35) ).arrange(DOWN, buff=0.35)
cta.move_to(ORIGIN)
self.play(Write(cta)) self.play(Write(cta))
self.wait(2.5) self.wait(2)
self.play(FadeOut(cta), FadeOut(title)) self.play(FadeOut(cta), FadeOut(title))