qwen3.5-27b-code-forged-mlx-4bit

continuum-ai/qwen3.5-27b-code-forged-mlx-4bit

qwen3.5-27b-code-forged-mlx-4bit is a text generation model from continuum-ai. Use it when you need the model to write or continue text. It is set up for mlx. The card lists the license as apache-2.0.

Qwen3.5-27B pruned by 30% and retrained for code through Experiential Plasticity.

Downloads · 30 days

226

5% of all-time downloads

All-time downloads

4.3K

Public

Parameters

14.3B

15.2 GB on disk

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.safetensors15.2 GB · 100%

Parameter types

How the weights are stored.

U813.4B · 94%

Try a prompt

Base models

Qwen/Qwen3.5-27B

Model card

From the Hugging Face model README

30% Smaller, +3.5% Better

Qwen3.5-27B pruned by 30% and retrained for code through Experiential Plasticity.

3.07 → 2.96 perplexity · 2 cycles

<a href="https://cambriantech.github.io/forge-alloy/verify/#hf.co/continuum-ai/qwen3.5-27b-code-forged-mlx-4bit/resolve/main/qwen3.5-27b-code-forged-mlx-4bit.alloy.json@6ca79c62b879cd4c"> <img src="alloy-qr.png" alt="Verify Chain of Custody" width="160"/> </a> <a href="https://cambriantech.github.io/forge-alloy/verify/#hf.co/continuum-ai/qwen3.5-27b-code-forged-mlx-4bit/resolve/main/qwen3.5-27b-code-forged-mlx-4bit.alloy.json@6ca79c62b879cd4c">Every claim on this card is verified</a> Trust: self-attested · 1 benchmark · 2 devices tested <a href="https://github.com/CambrianTech/forge-alloy">ForgeAlloy</a> chain of custody · <a href="qwen3.5-27b-code-forged-mlx-4bit.alloy.json">Download alloy</a> · Merkle-chained

Qwen3.5-27B with cryptographic provenance via the ForgeAlloy chain of custody.

Benchmarks

Benchmark	Result	Verified
perplexity	3.0	Self-reported

What Changed (Base → Forged)

	Base	Forged	Delta
Perplexity (code)	3.07	2.96	-3.5% ✅
Pruning	None	30% heads (magnitude)	-30% params ✅
Training	General	code, 500 steps	LR 2e-4, 2 cycles
Pipeline		prune → train → quant	2 cycles

Runs On

Device	Format	Size	Speed
MacBook Pro 32GB	fp16	—	Verified
RTX 3090 24GB	fp16	—	Verified
MacBook Pro 32GB	fp16	8.0GB	Expected
MacBook Air 16GB	Q8_0	~4.0GB	Expected
MacBook Air 8GB	Q4_K_M	~2.5GB	Expected
iPhone / Android	Q4_K_M	~2.5GB	Expected

Quick Start

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("continuum-ai/qwen3.5-27b-code-forged-mlx-4bit",
    torch_dtype="auto", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("continuum-ai/qwen3.5-27b-code-forged-mlx-4bit")

inputs = tokenizer("def merge_sort(arr):", return_tensors="pt").to(model.device)
output = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(output[0], skip_special_tokens=True))

Methodology

Produced via head pruning, GGUF quantization. Full methodology, ablations, and per-stage rationale are in the methodology paper and the companion MODEL_METHODOLOGY.md in this repository. The pipeline ran as prune → train → quant over 2 cycles on MacBook Pro 32GB.

Chain of Custody

Scan the QR or verify online. Download the alloy file to verify independently.

What	Proof
Model weights	`sha256:deff5ceac01ef81d2e6082bd14ac51fdc...`
Code that ran	`sha256:derivation-tool-o...`
Forged on	MacBook Pro 32GB, 2026-04-08
Trust level	`self-attested`
Spec	ForgeAlloy — Rust/Python/TypeScript

Make Your Own

Forged with Continuum — a distributed AI world that runs on your hardware.

The Factory configurator lets you design and forge custom models visually — context extension, pruning, LoRA, quantization, vision/audio modalities. Pick your target devices, the system figures out what fits.

GitHub · All Models · Forge-Alloy

License

apache-2.0

Task: Text Generation
Library: mlx
Type: qwen3_5
License: apache-2.0
Created: Mar 28, 2026
Updated: Apr 11, 2026