Arm/bge-base-en-v1.5-int8-litert

BAAI/bge-base-en-v1.5 optimized for Arm-based mobile CPUs with SME2

An INT8-quantized version of BAAI/bge-base-en-v1.5 for text embedding, exported to LiteRT (.tflite) and optimized for Arm-based mobile CPUs with SME2.

Summary

This repository contains an Arm-optimized version of BAAI/bge-base-en-v1.5 for text embedding, quantized to INT8 via post-training quantization — per-channel symmetric INT8 weights, with INT8 activations quantized dynamically at runtime. The model 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 34.646 ms (1.76x faster than baseline)
Accuracy result	99.35% cosine similarity against the FP32 baseline embeddings
Size / memory result	106.001 MB (3.91x smaller), peak memory 184.13 MB (3.62x less)

Original model

Field	Value
Original model	BAAI/bge-base-en-v1.5
Original source	Hugging Face
Original developer	BAAI (Beijing Academy of Artificial Intelligence)
Original model card	BAAI/bge-base-en-v1.5
Original license	MIT

Model files

File	Description
bge-base-en-v1.5_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	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
Execution backend	CPU (XNNPACK, KleidiAI), 1 thread
Precision	INT8, dynamic PTQ — per-channel symmetric weights, INT8 activations quantized dynamically at runtime
Batch size	1
Sequence length	128
Runs / warmup	100 / 20
CPU affinity	pinned to the 4.21 GHz ultra core

Performance results

Metric	Original / baseline	Arm-optimized	Improvement
Model size (MB)	414.77	106.001	3.91x smaller
End-to-end latency p50 (ms)	60.934	34.646	1.76x faster
End-to-end latency p90 (ms)	66.222	36.606	1.81x faster
Model load time (ms)	390.859	190.793	2.05x faster
Time to first inference (ms)	64.702	40.537	1.60x faster
Peak memory (MB)	667.3	184.13	3.62x less
Average memory (MB)	667.29	184.12	3.62x less
Requests per second	16.41	28.86	1.76x

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
Sample count	10000
Metric(s)	Cosine similarity against the FP32 baseline embeddings, embedding drift cosine distance
Runtime	LiteRT 0.9.0 (CPU, XNNPACK, KleidiAI)

Accuracy results

Metric	Original / baseline	Arm-optimized	Change
Cosine similarity (%)	100.0	99.35	-0.65 pp
Embedding drift cosine distance	0.0	0.0065	+0.0065

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 PTQ-dynamic — per-channel symmetric weights, INT8 activations quantized dynamically at runtime; no calibration data required
Runtime/backend selection	Yes	LiteRT CPU with XNNPACK and KleidiAI
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

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.

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

Expected input

Property	Value
Input shape	[1, 128]
Input type	int64
Input fields	input_ids [1, 128] int64, attention_mask [1, 128] int64
Preprocessing	tokenize with max_length 128, padding to max_length, truncation enabled

Expected output

Property	Value
Output shape	[1, 768]
Output type	float32
Output format	L2-normalized embedding vector
Postprocessing	mean pooling over non-masked tokens, then L2 normalization

Intended use

This model is intended for developers evaluating text embedding 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 10000 WikiText-2 samples using embedding similarity against the FP32 baseline, not a retrieval benchmark such as MTEB, and may not generalize to all domains.
• The input sequence length is fixed at 128 tokens; longer text is truncated.
• example.py downloads the BAAI/bge-base-en-v1.5 tokenizer from Hugging Face on first run, so the machine running the example needs network access to the Hugging Face Hub.
• 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

Quantization was performed with PT2E dynamic quantization. Every weight constant in the exported graph is INT8 with per-channel symmetric scales and a zero point of zero. Activations are INT8 as well, quantized dynamically rather than ahead of time, so no calibration dataset is used.

About this version

Original Model: BAAI/bge-base-en-v1.5 by BAAI (Beijing Academy of Artificial Intelligence) - 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 MIT.

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.