video-promo/architecture.py

126 lines
5.2 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

"""AstrAI promo: Full architecture overview — layer by layer introduction."""
from manim import *
class Architecture(Scene):
"""Reveals AstrAI's 5-layer inference stack, introducing each layer."""
def construct(self):
title = Text("AstrAI Architecture", font_size=44, color=BLUE)
self.play(Write(title))
self.play(title.animate.to_edge(UP, buff=0.3))
W = 8.0
def box(h=1.05, color=GRAY, fill=0.08):
return Rectangle(width=W, height=h, color=color, fill_opacity=fill, stroke_width=1.5)
def layer_header(text, color):
return Text(text, font_size=20, color=color, weight=BOLD)
def sub(text):
return Text(text, font_size=12, color=WHITE)
def intro(title, detail, color, oneline=None):
"""Animate a layer: box + title → details → brief pause."""
b = box(color=color, fill=0.1)
content = VGroup(title)
if oneline:
content.add(oneline)
if detail:
items = [title]
if oneline:
items.append(oneline)
items.extend(detail)
content = VGroup(*items)
else:
content = VGroup(title) if not oneline else VGroup(title, oneline)
content.arrange(DOWN, buff=0.15)
content.move_to(b.get_center())
grp = VGroup(b, content)
return grp
layers = []
# ── Layer 1: API Server ──
l1_t = layer_header("HTTP API Server", GREEN)
l1_d = [sub("FastAPI • OpenAI-Compatible /v1/chat/completions"),
sub("Streaming SSE • Async • Health/Stats Endpoints")]
l1 = intro(l1_t, l1_d, GREEN, sub("astrai/inference/server.py"))
l1.next_to(title, DOWN, buff=0.35)
layers.append(l1)
# ── Layer 2: Inference Engine ──
l2_t = layer_header("InferenceEngine", BLUE)
l2_d = [sub("generate() · generate_async() · generate_with_request()"),
sub("Batch mode · Streaming (Generator) · Thread-safe accumulator")]
l2 = intro(l2_t, l2_d, BLUE, sub("astrai/inference/engine.py"))
l2.next_to(l1, DOWN, buff=0.12)
layers.append(l2)
# ── Layer 3: Continuous Batching Scheduler ──
l3_t = layer_header("InferenceScheduler (Background Daemon)", YELLOW)
l3_d = [sub("Cleanup → Refill → Prefill → Decode · 4-phase loop"),
sub("Position-Grouped Decode · Bitmask O(1) Slot Allocation")]
l3 = intro(l3_t, l3_d, YELLOW, sub("astrai/inference/scheduler.py"))
l3.next_to(l2, DOWN, buff=0.12)
layers.append(l3)
# ── Layer 4: Prefix Cache + KV Cache ──
l4_t = layer_header("PrefixCacheManager + KV Cache", ORANGE)
l4_d = [sub("Radix Tree prefix matching · LRU eviction · Slot versioning"),
sub("GPU copy_() → Zero-Copy Reuse · k_cache / v_cache tensors")]
l4 = intro(l4_t, l4_d, ORANGE, sub("astrai/inference/scheduler.py"))
l4.next_to(l3, DOWN, buff=0.12)
layers.append(l4)
# ── Layer 5: Transformer Model ──
l5_t = layer_header("Transformer (1B params)", PURPLE)
l5_d = [sub("24× DecoderBlock · GQA 6:1 · RoPE · SwiGLU MLP"),
sub("Dim 1536 · Max Length 2048 · bfloat16 · 100K vocab")]
l5 = intro(l5_t, l5_d, PURPLE, sub("astrai/model/transformer.py"))
l5.next_to(l4, DOWN, buff=0.12)
layers.append(l5)
# ── Animate layer by layer ──
arrows = VGroup()
for i, layer in enumerate(layers):
self.play(Create(layer), run_time=0.4)
self.wait(1.0 if i < 2 else 0.8)
if i > 0:
prev = layers[i - 1][0]
curr = layer[0]
arrow = Arrow(
prev.get_bottom(), curr.get_top(),
color=GRAY, buff=0.06,
max_tip_length_to_length_ratio=0.18,
)
self.play(Create(arrow), run_time=0.15)
arrows.add(arrow)
self.wait(0.6)
# ── Highlight: the innovation layers ──
hl3 = SurroundingRectangle(layers[2], color=YELLOW, buff=0.1, stroke_width=2)
hl4 = SurroundingRectangle(layers[3], color=ORANGE, buff=0.1, stroke_width=2)
hl_note = Text("Key Innovations: Continuous Batching + Prefix Cache",
font_size=18, color=GOLD)
hl_note.next_to(VGroup(hl3, hl4), LEFT, buff=0.5)
self.play(Create(hl3), Create(hl4), Write(hl_note))
self.wait(2.0)
self.play(FadeOut(hl3), FadeOut(hl4), FadeOut(hl_note))
# ── Fade to CTA ──
self.play(FadeOut(VGroup(*layers)), FadeOut(arrows))
cta = VGroup(
Text("AstrAI", font_size=52, color=BLUE),
Text("Single GPU · Open Source · 1B params", font_size=22, color=GRAY),
Text("github.com/ViperEkura/AstrAI", font_size=18, color=YELLOW),
).arrange(DOWN, buff=0.35)
cta.move_to(ORIGIN)
self.play(Write(cta))
self.wait(2.5)
self.play(FadeOut(cta), FadeOut(title))