jica98 / qwen3_5_text

qwen3.5-4B-super-coder

Name: qwen3.5-4B-super-coder architecture and memory requirements
Creator: LLM Bottleneck
License: https://creativecommons.org/licenses/by/4.0/

Parameter count not published.

Architecturepublished data

Architecture available · weight sizes unavailable

Weight sizeno data

This snapshot holds qwen3.5-4B-super-coder’s architecture but no weight size for any format, so the calculator cannot answer for it and it has no hardware table.

Why: Neither a published artifact nor a published parameter count exists for this model at its pinned revision. A size is never inferred from the model’s name.

The coverage page explains what the catalogue does not size, and why.

Under the hood

Architecture, read from the publisher’s file

The numbers every figure above is computed from, with the file they came from.

Architecture

architecture.json · pinned 1f2362bff662

✓ Architecture read from the published config.json

Retrieved 2026-09-01 at pinned commit 1f2362bff662.

Architecture: qwen3_5_text
Layers: 32
Hidden size: 2,560
Attention heads: 16
KV heads: 4
Head dimension: 256
Feed-forward width: 9,216
Vocabulary: 248,320
Context ceiling: 262,144
RoPE theta: 10,000,000

Where the memory goes

Grouped-query attention shares each key/value head across 4 query heads, so the KV cache is 25% of what multi-head attention would need at the same context.

This is a multimodal checkpoint (vision); the diagram and memory sizing above cover the text decoder. Encoder/audio/image activations are not included in the KV or weight figures.

Open the exact official config.json ↗

Citing this page

LLM Bottleneck. “qwen3.5-4B-super-coder VRAM and hardware requirements.” Architecture from jica98/qwen3.5-4B-super-coder at revision 1f2362bff662, retrieved 2026-09-01. https://llmbottleneck.com/models/jica98-qwen3-5-4b-super-coder

Every figure above is either the published value or a reconstruction whose measured error is on the accuracy page.