Arm/minilm-l6-v2-int8-litert

all-MiniLM-L6-v2 optimized for Arm-based mobile CPUs with SME2

A sentence embedding model quantized to INT8 with dynamic post-training quantization and exported to LiteRT (.tflite) for Arm-based mobile CPUs with SME2.

Summary

This repository contains an Arm-optimized version of all-MiniLM-L6-v2 for feature extraction. The model maps text into 384-dimensional L2-normalized vectors for semantic similarity, retrieval, and feature extraction. It is provided in LiteRT (.tflite) format, targeting Mobile CPU systems.

This version is intended to demonstrate efficient inference on Arm-based platforms while preserving the original model's intended behavior. Arm has evaluated this model on WikiText-2 and measured performance on a representative evaluation target.

Key results

Area	Result
Model format	LiteRT .tflite
Target device class	Mobile CPU
Reference device	vivo X300 (C1-Ultra, C1-Premium, C1-Pro, Android 16 / OriginOS 6)
Primary performance result	p50 latency 7.332 ms (1.09x faster than baseline), 136.39 requests per second
Accuracy result	Cosine similarity 99.77% against the FP32 embeddings
Size / memory result	22.454 MB (3.82x smaller), peak memory 35.2 MB

Original model

Field	Value
Original model	sentence-transformers/all-MiniLM-L6-v2
Original source	Hugging Face
Original developer	Sentence Transformers
Original model card	sentence-transformers/all-MiniLM-L6-v2
Original license	Apache-2.0

Model files

File	Description
minilm-l6-v2_litert_optimized.tflite	Arm-optimized INT8 model for deployment
example.py	Minimal inference example
pyproject.toml	Pinned runtime dependencies for example.py, resolved with uv
uv.lock	Locked dependency resolution for pyproject.toml
config.yaml	Model I/O contract used by the example
benchmarks/	FP32 baseline and Arm-optimized benchmark records

Performance

Performance was measured on the reference configuration below. Results are intended to make the optimization reproducible but do not guarantee identical performance on every Arm-based system.

Reference configuration

Field	Value
Device / platform	vivo X300
CPU / accelerator	C1-Ultra, C1-Premium, C1-Pro, 8 cores @ 4.21 GHz
System memory	16 GB
OS	android, Android 16 / OriginOS 6
Runtime	LiteRT 0.9.0
Backend / delegate	XNNPACK, KleidiAI
Threads	1
Batch size	1
Sequence length	128
Precision	INT8, dynamic PTQ — per-channel symmetric weights, 8-bit activations
Runs	20 warmup + 100 measured

Performance results

Metric	Original / baseline	Arm-optimized	Improvement
End-to-end latency p50 (ms)	8.004	7.332	1.09x faster
End-to-end latency p90 (ms)	8.233	7.398	1.11x faster
Requests per second	124.94	136.39	1.09x
Model load time (ms)	67.882	47.146	1.44x faster
Time to first inference (ms)	13.526	9.518	1.42x faster
Model size (MB)	85.777	22.454	3.82x smaller
Peak memory (MB)	94.99	35.2	2.70x less
Average memory (MB)	94.99	35.19	2.70x less

Accuracy

Accuracy was evaluated using the same preprocessing, input resolution, and evaluation protocol described below. Where possible, the optimized model is compared against the original model under the same evaluation conditions.

Evaluation setup

Field	Value
Dataset	WikiText-2
Split	validation
Number of samples	10000
Metric(s)	Cosine similarity, embedding drift
Evaluation runtime	LiteRT 0.9.0

Accuracy results

Metric	Original / baseline	Arm-optimized	Change
Cosine similarity (%)	100.0	99.77	-0.23 pp
Embedding drift (cosine distance)	0.0	0.0023	+0.0023

Accuracy was measured using the evaluation setup described above. Users should re-evaluate the model on their own data before production use.

Arm optimization approach

Arm optimized this model for efficient inference on Arm-based platforms using a hardware-aware conversion and validation flow.

For this release, Arm used:

Optimization area	Applied?	Notes
Model conversion	Yes	Converted to LiteRT .tflite
Quantization	Yes	INT8 dynamic PTQ — per-channel symmetric weights, 8-bit activations
Runtime/backend selection	Yes	XNNPACK + KleidiAI delegate on CPU
Graph/runtime compatibility updates	Yes	Performed as part of the LiteRT export pipeline
Accuracy validation	Yes	Compared against the original model or published baseline
Performance validation	Yes	Measured on the reference Arm platform

The goal of this process is to improve deployment characteristics such as latency, memory use, model size, and runtime compatibility while preserving the model's intended behavior. Detailed conversion scripts, calibration configuration, or backend-specific implementation details may be provided separately where appropriate.

Using this model

Install dependencies

Dependencies are declared in pyproject.toml, which ships with this repository. Resolve and install them into a local virtual environment with uv:

uv python install
uv sync --frozen

Run the example

uv run example.py

Note: The Python/uv example runs on AWS Graviton (Ubuntu arm64) to confirm runtime compatibility only, and is intended as a guideline for building an equivalent run script on Mobile CPU systems.

Expected input

Property	Value
Input shape	input_ids [1, 128] and attention_mask [1, 128]
Input type	int64
Input range	Token ids from the all-MiniLM-L6-v2 WordPiece vocabulary
Preprocessing	preprocess() in example.py — tokenize with max_length 128, padding to max_length and truncation, cast to int64

Expected output

Property	Value
Output shape	[1, 384]
Output type	float32, L2-normalized embedding vector
Postprocessing	None — mean pooling and L2 normalization are applied inside the model wrapper

Intended use

This model is intended for developers evaluating feature extraction workloads on Arm-based platforms. It is suitable as a reference implementation for benchmarking, prototyping, and integration exploration.

Limitations

• Performance depends on the target device, runtime version, backend/delegate support, memory configuration, and system load.
• Accuracy was evaluated on the WikiText-2 validation split and may not generalize to all domains.
• This release preserves the original model's intended task and behavior, but users should validate it for their own application, data, and deployment environment.
• This repository is not a replacement for the original model documentation.

Additional notes

Inputs longer than 128 tokens are truncated, and shorter inputs are padded to 128 — the exported graph has a fixed sequence length.

example.py downloads the all-MiniLM-L6-v2 tokenizer from Hugging Face on first run, so the machine running it needs network access. The tokenizer is not bundled in this repository.

Runtime metrics were measured single-threaded, pinned to the 4.21 GHz ultra core of the vivo X300, over 20 warmup and 100 measured runs.

About this version

Original Model: sentence-transformers/all-MiniLM-L6-v2 by Sentence Transformers - Repository

Optimization/conversion: Arm-Optimized version for execution on Arm-based platforms.

Converted/optimized by: Arm

License: The Original Model and the Optimized Model are subject to Apache-2.0.

This repository contains a converted or optimized version of the Original Model (the “Optimized Model”). The Original Model has been converted or optimized as described above for execution on Arm-based platforms.

No retraining or fine-tuning of the Original Model was performed as part of the conversion or optimization. The conversion or optimization was not intended to change the Original Model’s behavior or intended use.

Original Model and Documentation

For information about the Original Model, including its development, training data, intended uses, limitations and other relevant information, please refer to the Original Model repository. Information in that repository was provided by the original developer or other third parties and, unless expressly stated otherwise, has not been independently verified by Arm.

Licenses and Third-Party Terms

Use of the Original Model and the Optimized Model is subject to the applicable licenses, usage restrictions and other terms identified above and in the relevant repositories. Publication of the Optimized Model does not grant any rights beyond those provided under the applicable license terms.

You are responsible for reviewing those terms and ensuring that your use of the Original Model and the Optimized Model is permitted.

Purpose of this Release

The Optimized Model is provided as a reference implementation to demonstrate and evaluate execution and performance on Arm-based systems. It is not a production-ready or supported solution.

Arm’s publication of the Optimized Model does not constitute an endorsement or certification of the Original Model or a representation that the Optimized Model is suitable for production use or any particular purpose.

To the fullest extent permitted by applicable law (i) the Optimized Model is provided “as is.” Arm makes no representations or warranties that the Original Model, the Optimized Model or their outputs are accurate, safe, secure, non-infringing, legally compliant, suitable for production use or fit for any particular purpose; and (ii) Arm will not be liable for any loss or damage arising from or in connection with the Optimized Model, its use or its outputs.

You are responsible for independently evaluating the Optimized Model, its outputs and its suitability for your intended use, including compliance with applicable legal, regulatory, safety and security requirements.

Arm does not commit to provide ongoing support, maintenance or updates for the Optimized Model. Any use of or reliance on the Optimized Model or its outputs is at your own risk.