Spaces:

SupraLabs
/

Blog

Running

App Files Files Community

AxionLab-official commited on 8 days ago

Commit

408c06e

verified ·

1 Parent(s): 872d88a

Create supra-v5-8m.html

Browse files

Files changed (1) hide show

supra-v5-8m.html +375 -0

supra-v5-8m.html ADDED Viewed

	@@ -0,0 +1,375 @@

+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>Supra Mini v5 8M | SupraLabs Blog</title>
+    <style>
+        :root {
+            --bg: #0f0f0f;
+            --surface: #1a1a1a;
+            --border: #333;
+            --text: #e0e0e0;
+            --accent: #536bfe;
+            --muted: #888;
+            --font-mono: 'JetBrains Mono', 'Fira Code', monospace;
+        }
+        * { margin: 0; padding: 0; box-sizing: border-box; }
+        body {
+            background-color: var(--bg);
+            color: var(--text);
+            font-family: 'Inter', -apple-system, sans-serif;
+            line-height: 1.6;
+            padding: 2rem;
+        }
+        code, pre, .mono { font-family: var(--font-mono); }
+        .container { max-width: 900px; margin: 0 auto; }
+        header {
+            border-bottom: 2px solid var(--border);
+            padding-bottom: 2rem;
+            margin-bottom: 3rem;
+            display: flex;
+            justify-content: space-between;
+            align-items: flex-end;
+        }
+        .logo-area h1 {
+            font-size: 1.2rem;
+            text-transform: uppercase;
+            letter-spacing: 2px;
+            color: var(--accent);
+            line-height: 1;
+            display: flex;
+            align-items: center;
+            gap: 10px;
+        }
+        .logo-area a { text-decoration: none; color: inherit; }
+        .logo-area {
+            display: flex;
+            align-items: center;
+            gap: 10px;
+            font-weight: bold;
+            font-size: 1.2rem;
+        }
+        nav a {
+            color: var(--text);
+            text-decoration: none;
+            margin-left: 1.5rem;
+            font-size: 0.9rem;
+            border-bottom: 1px solid transparent;
+        }
+        nav a:hover { border-bottom: 1px solid var(--accent); }
+        .post-header { margin-bottom: 3rem; }
+        .post-header h2 {
+            font-size: 3rem;
+            line-height: 1.1;
+            margin-bottom: 1rem;
+            font-weight: 800;
+        }
+        .post-meta {
+            font-family: var(--font-mono);
+            color: var(--accent);
+            font-size: 0.9rem;
+            margin-bottom: 2rem;
+        }
+        .post-content {
+            background: var(--surface);
+            border: 1px solid var(--border);
+            padding: 3rem;
+            margin-bottom: 4rem;
+        }
+        .post-content h2 {
+            font-size: 1.8rem;
+            margin: 2.5rem 0 1rem 0;
+            color: var(--accent);
+        }
+        .post-content h2:first-child { margin-top: 0; }
+        .post-content p {
+            margin-bottom: 1.5rem;
+            font-size: 1.1rem;
+            color: var(--text);
+        }
+        .post-content ul {
+            margin-bottom: 1.5rem;
+            padding-left: 1.5rem;
+        }
+        .post-content li { margin-bottom: 0.5rem; font-size: 1.1rem; }
+        .post-content strong { color: #fff; }
+        .post-content code {
+            background: #111;
+            border: 1px solid var(--border);
+            padding: 2px 6px;
+            border-radius: 3px;
+            font-size: 0.95em;
+            color: var(--accent);
+        }
+        .code-block {
+            background: #111;
+            border: 1px solid var(--border);
+            padding: 1.5rem;
+            margin: 2rem 0;
+            overflow-x: auto;
+            font-family: var(--font-mono);
+            font-size: 0.88rem;
+            line-height: 1.7;
+            color: #ccc;
+        }
+        .code-block .comment { color: var(--muted); }
+        .code-block .keyword { color: var(--accent); }
+        .callout {
+            border-left: 3px solid var(--accent);
+            background: #111;
+            padding: 1rem 1.5rem;
+            margin: 2rem 0;
+            font-family: var(--font-mono);
+            font-size: 0.95rem;
+            color: #ccc;
+        }
+        .callout span {
+            display: block;
+            color: var(--muted);
+            font-size: 0.8rem;
+            margin-bottom: 0.4rem;
+        }
+        .output-example {
+            border: 1px solid var(--border);
+            background: #111;
+            padding: 1.5rem;
+            margin: 1.5rem 0;
+        }
+        .output-example .prompt-label {
+            font-family: var(--font-mono);
+            font-size: 0.75rem;
+            color: var(--accent);
+            margin-bottom: 0.4rem;
+        }
+        .output-example .prompt-text {
+            font-weight: 700;
+            color: #fff;
+            margin-bottom: 1rem;
+            font-size: 1rem;
+        }
+        .output-example .output-label {
+            font-family: var(--font-mono);
+            font-size: 0.75rem;
+            color: var(--muted);
+            margin-bottom: 0.4rem;
+        }
+        .output-example .output-text {
+            color: var(--text);
+            font-size: 0.95rem;
+            font-style: italic;
+            line-height: 1.7;
+        }
+        .table-wrap { overflow-x: auto; margin: 2rem 0; }
+        table {
+            width: 100%;
+            border-collapse: collapse;
+            font-family: var(--font-mono);
+            font-size: 0.9rem;
+        }
+        th {
+            background: #111;
+            color: var(--accent);
+            padding: 0.75rem 1rem;
+            text-align: left;
+            border: 1px solid var(--border);
+        }
+        td {
+            padding: 0.7rem 1rem;
+            border: 1px solid var(--border);
+            color: var(--text);
+        }
+        tr:nth-child(even) td { background: #111; }
+        .tags { display: flex; gap: 0.5rem; margin-top: 2rem; flex-wrap: wrap; }
+        .tag {
+            font-family: var(--font-mono);
+            font-size: 0.7rem;
+            padding: 2px 8px;
+            border: 1px solid var(--border);
+            border-radius: 4px;
+            color: var(--muted);
+        }
+        footer {
+            margin-top: 6rem;
+            padding-bottom: 2rem;
+            font-size: 0.8rem;
+            color: var(--muted);
+            text-align: center;
+        }
+        @media (max-width: 600px) {
+            .post-header h2 { font-size: 2rem; }
+            .post-content { padding: 1.5rem; }
+            header { flex-direction: column; align-items: flex-start; gap: 1rem; }
+            nav a { margin-left: 0; margin-right: 1rem; }
+        }
+    </style>
+</head>
+<body>
+    <div class="container">
+        <header>
+            <div class="logo-area" style="font-size: 1.5em;">
+                <a href="./index.html"><h1><img src="./image.png" style="height: 2em"> SupraLabs_</h1></a>
+            </div>
+            <nav>
+                <a href="./index.html#news">News</a>
+                <a href="https://huggingface.co/SupraLabs" target="blank">HuggingFace</a>
+                <a href="./index.html#hardware">Hardware</a>
+            </nav>
+        </header>
+        <article>
+            <div class="post-header">
+                <div class="post-meta">// 2026-05-16 | Release</div>
+                <h2>🦅 Supra Mini v5 8M<br>just dropped.</h2>
+            </div>
+            <div class="post-content">
+                <p>We are back. <strong>Supra Mini v5 8M</strong> is our fifth release in the Supra Mini series and the biggest one yet. Trained on <strong>5 billion tokens</strong> of Fineweb-Edu for 2 epochs, v5 jumps to 7.8M parameters and brings real, measurable improvements across every benchmark we ran.</p>
+                <h2>What changed from v4?</h2>
+                <p>v4 had 2.6M parameters and was trained on 3B tokens. v5 triples the parameter count to <strong>7.8M</strong> and adds 2 billion more tokens to the training set. The vocab size also doubled from 8,192 to <strong>16,384</strong>, giving the model a much richer token vocabulary to work with. Every single part of the config got an upgrade.</p>
+                <div class="callout">
+                    <span>// supra mini v5 8m — model config</span>
+                    Parameters &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;→ 7,867,584 (~8M)<br>
+                    Architecture &nbsp;&nbsp;&nbsp;→ Llama<br>
+                    Vocab size &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;→ 16,384 (custom BPE)<br>
+                    Hidden size &nbsp;&nbsp;&nbsp;&nbsp;→ 191<br>
+                    Intermediate &nbsp;&nbsp;→ 768<br>
+                    Layers &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;→ 8<br>
+                    Attention heads &nbsp;→ 4<br>
+                    Context length &nbsp;&nbsp;→ 1,024 tokens<br>
+                    Learning rate &nbsp;&nbsp;→ 2e-4<br>
+                    Weight decay &nbsp;&nbsp;&nbsp;→ 0.01<br>
+                    Trained in &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;→ bfloat16
+                </div>
+                <h2>Training setup</h2>
+                <p>We trained v5 on a <strong>single NVIDIA RTX 5060 Ti 16GB</strong> for approximately 11 hours across 2 epochs. The dataset is the first 5 billion tokens of Sample-10BT from <strong>Fineweb-Edu</strong>, streamed and tokenized on the fly with our custom BPE tokenizer (vocab size 16,384).</p>
+                <p>The final training loss after 2 epochs came in at <strong>4.414</strong>, down from 4.618 in v4. The full training code is in the repo: <code>train_tokenizer.py</code>, <code>train.py</code> and <code>inference.py</code>.</p>
+                <h2>Benchmarks</h2>
+                <p>We ran the same <code>lm-eval</code> suite as v4 so the numbers are directly comparable. v5 beats v4 on every single task.</p>
+                <div class="table-wrap">
+                    <table>
+                        <thead>
+                            <tr>
+                                <th>Task</th>
+                                <th>v5 score</th>
+                                <th>v4 score</th>
+                                <th>Random baseline</th>
+                            </tr>
+                        </thead>
+                        <tbody>
+                            <tr><td>ARC_Easy</td><td>0.3439</td><td>0.3152</td><td>0.25 (25%)</td></tr>
+                            <tr><td>Wikitext (PPL)</td><td>2.6617</td><td>3.1652</td><td>lower is better</td></tr>
+                            <tr><td>BLiMP</td><td>0.6349</td><td>0.607</td><td>0.50 (50%)</td></tr>
+                        </tbody>
+                    </table>
+                </div>
+                <p>The Wikitext perplexity drop from 3.16 to 2.66 is the one that stands out most. Lower perplexity means the model is genuinely more confident and accurate at predicting text, not just getting lucky on multiple choice. BLiMP going from 0.607 to 0.634 also shows real grammatical improvement for a model this size.</p>
+                <h2>Example outputs</h2>
+                <p>Same prompts as v4, so you can compare directly. Generation settings: <code>temperature=0.5</code>, <code>top_k=25</code>, <code>top_p=0.9</code>.</p>
+                <div class="output-example">
+                    <div class="prompt-label">// prompt</div>
+                    <div class="prompt-text">"Artificial intelligence is "</div>
+                    <div class="output-label">// output</div>
+                    <div class="output-text">"Artificial intelligence is _______. I was a scientist at the University of Cambridge in London, and I had been looking for ways to get them back into the first time. It's not just about learning how to use it but also about the world around us. The study has led to more than 150 studies showing that people who are able to learn new things like computers, laptops, and smartphones can be found in many different types of computer systems..."</div>
+                </div>
+                <div class="output-example">
+                    <div class="prompt-label">// prompt</div>
+                    <div class="prompt-text">"The main concept of physics is "</div>
+                    <div class="output-label">// output</div>
+                    <div class="output-text">"The main concept of physics is _______. This principle, which means that the universe has no energy; it should be a matter of time in its own right and therefore to be able to achieve its maximum potential. It must have been used as an alternative for quantum mechanics but not only in terms of power but also in the form of electrical systems such as magnetic fields or electric currents..."</div>
+                </div>
+                <div class="output-example">
+                    <div class="prompt-label">// prompt</div>
+                    <div class="prompt-text">"Once upon a time, "</div>
+                    <div class="output-label">// output</div>
+                    <div class="output-text">"Once upon a time, .. ... The word 'pot' is not used in the past. It's like a thing of course that we're going to think about it. The phrase 'these' means what you see and how much are there? You have no idea where I'm so, but if my son has any thought or understanding of his name, he will be able to understand him by saying something more than one day..."</div>
+                </div>
+                <p>The outputs are noticeably more coherent than v4. Sentences are longer, the topic stays more consistent, and the hallucinations are less random. Still a base model doing base model things, but the progress is clear.</p>
+                <h2>How to run it</h2>
+                <div class="code-block">
+<span class="keyword">from</span> transformers <span class="keyword">import</span> pipeline<br>
+<span class="keyword">import</span> torch<br>
+<br>
+<span class="keyword">print</span>(<span class="string">"[*] Loading Supra Mini v5 8M model from Hugging Face Hub..."</span>)<br>
+pipe = pipeline(<br>
+&nbsp;&nbsp;&nbsp;&nbsp;<span class="string">"text-generation"</span>,<br>
+&nbsp;&nbsp;&nbsp;&nbsp;model=<span class="string">"SupraLabs/Supra-Mini-v5-8M"</span>,<br>
+&nbsp;&nbsp;&nbsp;&nbsp;device_map=<span class="string">"auto"</span>,<br>
+&nbsp;&nbsp;&nbsp;&nbsp;torch_dtype=torch.float16 <span class="keyword">if</span> torch.cuda.is_available() <span class="keyword">else</span> torch.float32<br>
+)<br>
+<br>
+<span class="keyword">def</span> generate_text(prompt, max_length=150):<br>
+&nbsp;&nbsp;&nbsp;&nbsp;result = pipe(<br>
+&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;prompt,<br>
+&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;max_new_tokens=max_length,<br>
+&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;do_sample=True,<br>
+&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;temperature=0.5,<br>
+&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;top_k=25,<br>
+&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;top_p=0.9,<br>
+&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;repetition_penalty=1.2,<br>
+&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;pad_token_id=pipe.tokenizer.pad_token_id,<br>
+&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;eos_token_id=pipe.tokenizer.eos_token_id<br>
+&nbsp;&nbsp;&nbsp;&nbsp;)<br>
+&nbsp;&nbsp;&nbsp;&nbsp;<span class="keyword">return</span> result[0][<span class="string">'generated_text'</span>]<br>
+<br>
+test_prompt = <span class="string">"The importance of education is"</span><br>
+<span class="keyword">print</span>(f<span class="string">"\nPrompt: {test_prompt}"</span>)<br>
+<span class="keyword">print</span>(<span class="string">"-" * 30</span>)<br>
+<span class="keyword">print</span>(<span class="string">"\nOutput:\n"</span> + generate_text(test_prompt))
+                </div>
+                <h2>What's next?</h2>
+                <p>v5 is still a base model. No instruction tuning, no chat format, just raw language modeling. The roadmap for the next versions includes fine-tuning experiments and continuing to scale up parameters while keeping everything trainable on consumer hardware.</p>
+                <p><strong>The model is live. Go try it.</strong></p>
+                <div class="callout">
+                    <span>// links</span>
+                    Model &nbsp;&nbsp;→ huggingface.co/SupraLabs/Supra-Mini-v5-8M<br>
+                    License → Apache 2.0<br>
+                    Series &nbsp;→ Supra Mini collection on HuggingFace
+                </div>
+                <div class="tags">
+                    <span class="tag">#release</span>
+                    <span class="tag">#supra-mini-v5</span>
+                    <span class="tag">#tinyml</span>
+                    <span class="tag">#llama</span>
+                    <span class="tag">#open-source</span>
+                    <span class="tag">#fineweb-edu</span>
+                    <span class="tag">#edge-ai</span>
+                </div>
+            </div>
+        </article>
+        <footer>
+            <p class="mono">&copy; 2026 SupraLabs // Built for the community.</p>
+        </footer>
+    </div>
+</body>
+</html>