WEBGEN-4B-Preview-imatrix-GGUF

duyntnet/WEBGEN-4B-Preview-imatrix-GGUF

WEBGEN-4B-Preview-imatrix-GGUF is a text generation model from duyntnet. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as other.

Quantizations of https://huggingface.co/Tesslate/WEBGEN-4B-Preview

Downloads · 30 days

479

9% of all-time downloads

All-time downloads

5.5K

Public

Repo size

58.2 GB

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.gguf58.2 GB · 100%

At a glance

Task: Text Generation
Library: transformers
License: other
Access: Public
Created: Sep 3, 2025
Updated: Sep 3, 2025
SHA: df2ed013

Try a prompt

Model card

From the Hugging Face model README

Quantizations of https://huggingface.co/Tesslate/WEBGEN-4B-Preview

Open source inference clients/UIs

Closed source inference clients/UIs

LM Studio
More will be added...

From original readme

Quickstart

Transformers

<div class="codeblock"><pre> from transformers import AutoModelForCausalLM, AutoTokenizer import torch

model_id = "Tesslate/WEBGEN-4B-Preview" tok = AutoTokenizer.from_pretrained(model_id) model = AutoModelForCausalLM.from_pretrained( model_id, torch_dtype=torch.bfloat16, device_map="auto" )

prompt = """Make a single-file landing page for 'LatticeDB'. Style: modern, generous whitespace, Tailwind, rounded-xl, soft gradients. Sections: navbar, hero (headline + 2 CTAs), features grid, pricing (3 tiers), FAQ accordion, footer. Constraints: semantic HTML, no external JS."""

inputs = tok(prompt, return_tensors="pt").to(model.device) out = model.generate(**inputs, max_new_tokens=2000, temperature=0.7, top_p=0.9) print(tok.decode(out[0], skip_special_tokens=True)) </pre></div>

vLLM

<div class="codeblock"><pre> vllm serve Tesslate/WEBGEN-4B-Preview \ --host 0.0.0.0 --port 8000 \ --max-model-len 65536 \ --gpu-memory-utilization 0.92 </pre></div>

sglang

<div class="codeblock"><pre> python -m sglang.launch_server \ --model-path Tesslate/WEBGEN-4B-Preview \ --host 0.0.0.0 --port 5000 \ --mem-fraction-static 0.94 \ --attention-backend flashinfer \ --served-model-name webgen-4b </pre></div>

Tip: Lower temperature (e.g., 0.4–0.6) yields stricter, cleaner markup. Raise it for more visual variety.

<hr/>

Inference Settings (suggested)

<table> <thead><tr><th>Param</th><th>Value</th><th>Notes</th></tr></thead> <tbody> <tr><td><code>temperature</code></td><td>0.6</td><td>Balance creativity & consistency (lower if quantized)</td></tr> <tr><td><code>top_p</code></td><td>0.9</td><td>Nucleus sampling</td></tr> <tr><td><code>top_k</code></td><td>40</td><td>Optional vocab restriction</td></tr> <tr><td><code>max_new_tokens</code></td><td>1200–2500</td><td>Single-file sites often fit < 1500</td></tr> <tr><td><code>repetition_penalty</code></td><td>1.1</td><td>Reduces repetitive classes/markup</td></tr> </tbody> </table> <hr/>

Prompts that work well

<div class="grid grid-2"> <div class="card"> <h3>Starter</h3> <div class="codeblock"><pre> Make a single-file landing page for "RasterFlow" (GPU video pipeline). Style: modern tech, muted palette, Tailwind, rounded-xl, subtle gradients. Sections: navbar, hero (big headline + 2 CTAs), logos row, features (3x cards), code block (copyable), pricing (3 tiers), FAQ accordion, footer. Constraints: semantic HTML, no external JS. Return ONLY the HTML code. </pre></div> </div> <div class="card"> <h3>Design-strict</h3> <div class="codeblock"><pre> Use an 8pt spacing system. Palette: slate with indigo accents. Typography scale: 14/16/18/24/36/56. Max width: 1200px. Avoid shadows > md; prefer borders/dividers. </pre></div> </div> </div> <hr/>

Quantization & VRAM (example)

<table> <thead><tr><th>Format</th><th>Footprint</th><th>Notes</th></tr></thead> <tbody> <tr><td>BF16</td><td>8.05 GB</td><td>Fastest, best fidelity</td></tr> <tr><td>GGUF Q5_K_M</td><td>2.89 GB</td><td>Great quality/size trade-off</td></tr> <tr><td>GGUF Q4_K_M</td><td>2.5 GB</td><td>Smallest comfortable for laptops</td></tr> </tbody> </table> <hr/>

Intended Use & Scope

Primary: Generate complete, single-file websites (landing pages, marketing pages, simple docs) with semantic HTML and Tailwind classes.
Secondary: Component blocks (hero, pricing, FAQ) for manual composition.

<details> <summary><strong>Limitations</strong></summary> <ul> <li>Accessibility: adds headings/labels but ARIA coverage may need review.</li> <li>JS widgets: kept light unless explicitly requested in prompt.</li> </ul> </details> <details> <summary><strong>Ethical Considerations</strong></summary> <ul> <li>Curate prompts appropriately.</li> <li>When using third-party logos/assets, ensure you have rights or use open sources.</li> </ul> </details> <hr/>

Training Summary (research preview)

Base: <code>Qwen/Qwen3-4B-Instruct</code>
Objective: Tight web-only bias; reward semantic structure, spacing rhythm, and responsiveness.
Data: Mixture of curated HTML/CSS/Tailwind snippets, component libraries, and synthetic page specs.
Recipe: SFT with format constraints → instruction tuning → style/rhythm preference optimization.
Context: effective ~64k; trained to keep default outputs within practical page length.

<hr/>

Example Outputs

Community

Examples: <a href="https://uigenoutput.tesslate.com">uigenoutput.tesslate.com</a>
Discord: <a href="https://discord.gg/EcCpcTv93U">discord.gg/EcCpcTv93U</a>
Website: <a href="https://tesslate.com">tesslate.com</a>

<blockquote> “Why are good design models so expensive” — Tesslate Team </blockquote> <hr/>

Task: Text Generation
Library: transformers
License: other
Languages: en
Created: Sep 3, 2025
Updated: Sep 3, 2025