anemll-qwen3-4b-2507-ctx2048 – AI Model by Intension | AlphaNeural AI

You can deploy this model and start earning money today!

Intension/
anemll-qwen3-4b-2507-ctx2048

like

0

apple-neural-engine

Qwen/Qwen3-4B-Instruct-2507

Views

No views yet

anemll-qwen3-4b-2507-ctx2048

Qwen3-4B-Instruct-2507 converted to Core ML for the Apple Neural Engine using ANEMLL v0.3.5.

This artifact powers on-device session classification in Intension. It is downloaded automatically by the app; there is nothing to install manually.

Conversion parameters

Parameter	Value
Base model	Qwen/Qwen3-4B-Instruct-2507
Context length	2048
Batch size (prefill)	64
FFN / LM-head quantization	LUT6, per-channel 4
Chunks	4
ANEMLL version	0.3.5

See meta.yaml for the full conversion record.

Layout

qwen_embeddings.mlmodelc — embeddings
qwen_FFN_PF_lut6_chunk_0Xof04.mlmodelc — FFN + prefill chunks
qwen_lm_head_lut6.mlmodelc — LM head
meta.yaml — model/conversion metadata consumed by the runtime
Tokenizer files (tokenizer.json, tokenizer_config.json, vocab.json, merges.txt)

Licenses & attribution

Base model weights: Qwen/Qwen3-4B-Instruct-2507, Apache-2.0 © Alibaba Cloud.
Conversion pipeline: ANEMLL, MIT.
This repository redistributes a derivative (quantized, Core ML-compiled) form of the base model under Apache-2.0.