Activity v1
Tokenization and Context Window Visualizer
Inspect deterministic token approximations and allocate a finite context budget.
completion
beginner
seed 11Tokenization and Context Window Visualizer: Baseline
Explore the core controls with stable inputs and visible assumptions.
Assumptions
The simulation is deterministic and intentionally simplifies provider and hardware behavior.
Failure injection
Stable baseline with no injected production fault.
Inference Kernel Simulator
RadixAttention + PagedAttention v2
Token Context Window & Prefix Cache Explorer
Explore how shared prompt prefix caching and paged KV memory tables eliminate redundant VRAM allocations.
Prefix Cache Hit44%
VRAM Blocks Used120 / 128
1 seq4 seqs8 seqs
128 tok512 tok1024 tok
RadixTree Prefix Caching
Active: Shared system prompt tokens are computed once and reused across all concurrent requests.
GPU Memory Pressure94%
Radix Tree Prefix Hierarchy
Shared Root Node
Root Prefix
512 tokens
32 Physical Blocks
Seq #1
+128 prompt
+64 gen
Total: 704 tok
Seq #2
+256 prompt
+96 gen
Total: 864 tok
Seq #3
+192 prompt
+128 gen
Total: 832 tok
Seq #4
+384 prompt
+160 gen
Total: 1056 tok
Physical GPU KV Block Table (16 Tokens / Block)
96 blocks (1536 tokens) saved via zero-copy0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
RadixTree prefix sharing reduced VRAM consumption by 1536 tokens across 4 streams.
Zero-Copy Verified
0 saved attempts
Expected outcomes
- Estimate context consumption
- Recognize token boundary effects