Skip to content

Remove svg on int8 kv cache attn #252

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
kunal-mansukhani merged 1 commit into from
Apr 19, 2026
Merged

Remove svg on int8 kv cache attn #252

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
60 changes: 0 additions & 60 deletions challenges/medium/96_int8_kv_cache_attention/challenge.html
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,66 +9,6 @@
<code>float32</code>.
</p>

<svg width="700" height="270" viewBox="0 0 700 270" xmlns="http://www.w3.org/2000/svg" style="display:block; margin:20px auto;">
<rect width="700" height="270" fill="#222" rx="10"/>
<text x="350" y="26" fill="#ccc" font-family="monospace" font-size="13" text-anchor="middle">INT8 KV-Cache Attention — single token decode</text>

<!-- Q box -->
<rect x="20" y="45" width="80" height="36" fill="#2563eb" rx="4"/>
<text x="60" y="68" fill="#fff" font-family="monospace" font-size="12" text-anchor="middle">Q (fp32)</text>

<!-- K_int8 + scale -->
<rect x="130" y="45" width="100" height="36" fill="#7c3aed" rx="4"/>
<text x="180" y="63" fill="#fff" font-family="monospace" font-size="11" text-anchor="middle">K_int8 (int8)</text>
<rect x="130" y="90" width="100" height="28" fill="#5b21b6" rx="4"/>
<text x="180" y="109" fill="#fff" font-family="monospace" font-size="11" text-anchor="middle">k_scale (fp32)</text>

<!-- V_int8 + scale -->
<rect x="260" y="45" width="100" height="36" fill="#065f46" rx="4"/>
<text x="310" y="63" fill="#fff" font-family="monospace" font-size="11" text-anchor="middle">V_int8 (int8)</text>
<rect x="260" y="90" width="100" height="28" fill="#064e3b" rx="4"/>
<text x="310" y="109" fill="#fff" font-family="monospace" font-size="11" text-anchor="middle">v_scale (fp32)</text>

<!-- dequant arrows -->
<line x1="180" y1="118" x2="180" y2="148" stroke="#a78bfa" stroke-width="1.5" marker-end="url(#arr2)"/>
<line x1="310" y1="118" x2="310" y2="148" stroke="#34d399" stroke-width="1.5" marker-end="url(#arr2)"/>
<text x="180" y="145" fill="#a78bfa" font-family="monospace" font-size="10" text-anchor="middle">×</text>
<text x="310" y="145" fill="#34d399" font-family="monospace" font-size="10" text-anchor="middle">×</text>

<!-- K_float, V_float -->
<rect x="130" y="155" width="100" height="28" fill="#4c1d95" rx="4"/>
<text x="180" y="174" fill="#e9d5ff" font-family="monospace" font-size="11" text-anchor="middle">K_float</text>
<rect x="260" y="155" width="100" height="28" fill="#022c22" rx="4"/>
<text x="310" y="174" fill="#6ee7b7" font-family="monospace" font-size="11" text-anchor="middle">V_float</text>

<!-- attention flow -->
<line x1="60" y1="81" x2="60" y2="210" stroke="#60a5fa" stroke-width="1.5"/>
<line x1="60" y1="210" x2="160" y2="210" stroke="#60a5fa" stroke-width="1.5" marker-end="url(#arr2)"/>
<line x1="180" y1="183" x2="180" y2="210" stroke="#a78bfa" stroke-width="1.5" marker-end="url(#arr2)"/>
<rect x="155" y="205" width="50" height="26" fill="#1e1b4b" rx="4"/>
<text x="180" y="222" fill="#c4b5fd" font-family="monospace" font-size="10" text-anchor="middle">scores</text>
<line x1="205" y1="218" x2="250" y2="218" stroke="#e5e7eb" stroke-width="1.5" marker-end="url(#arr2)"/>
<rect x="250" y="205" width="70" height="26" fill="#1e293b" rx="4"/>
<text x="285" y="222" fill="#93c5fd" font-family="monospace" font-size="10" text-anchor="middle">softmax</text>
<line x1="320" y1="218" x2="360" y2="218" stroke="#e5e7eb" stroke-width="1.5" marker-end="url(#arr2)"/>
<line x1="310" y1="183" x2="370" y2="218" stroke="#34d399" stroke-width="1.5"/>
<rect x="360" y="205" width="60" height="26" fill="#1a2e1a" rx="4"/>
<text x="390" y="222" fill="#86efac" font-family="monospace" font-size="10" text-anchor="middle">output</text>

<!-- formula text -->
<text x="470" y="75" fill="#93c5fd" font-family="monospace" font-size="11">K[h,s,:] = K_int8[h,s,:] × k_scale[h,s]</text>
<text x="470" y="95" fill="#6ee7b7" font-family="monospace" font-size="11">V[h,s,:] = V_int8[h,s,:] × v_scale[h,s]</text>
<text x="470" y="125" fill="#fde68a" font-family="monospace" font-size="11">scores[h,s] = Q[h,:]·K[h,s,:] / √head_dim</text>
<text x="470" y="145" fill="#fde68a" font-family="monospace" font-size="11">w[h,:] = softmax(scores[h,:])</text>
<text x="470" y="165" fill="#fde68a" font-family="monospace" font-size="11">out[h,:] = Σ_s w[h,s] · V[h,s,:]</text>

<defs>
<marker id="arr2" markerWidth="6" markerHeight="6" refX="3" refY="3" orient="auto">
<path d="M0,0 L0,6 L6,3 z" fill="#888"/>
</marker>
</defs>
</svg>

<h2>Implementation Requirements</h2>
<ul>
<li>Implement the function <code>solve(Q, K_int8, V_int8, k_scale, v_scale, output, num_heads, seq_len, head_dim)</code>.</li>
Expand Down
Loading