Mo-Abdelfattah

arabic-semantic-chunker-qwen1.5b

README

License: apache-2.0

Model Overview
Intended Use
Training Details
Dataset Architecture
Infrastructure & Hardware
Load Testing & Performance
Usage Guide
Output Schema
Limitations & Boundary Conditions

Model Overview

Table with columns: Attribute, Value
Attribute	Value
Base Model	`Qwen/Qwen2.5-1.5B-Instruct`
Task	Text Segmentation / Multilingual Semantic Chunking
Finetuning Method	Supervised Fine-Tuning (SFT) with LoRA
Framework Stack	Unsloth + Hugging Face TRL (`SFTTrainer`)
Supported Languages	Arabic, English, and Code-Switched (Mixed) Text
License	Apache 2.0

This model acts as a dedicated LoRA Adapter layered over Qwen2.5-1.5B-Instruct. When fed a text corpus, it splits it into logically self-contained, semantically coherent chunks. It preserves 100% of the input text—no paraphrasing, translation, summarization, or omission occurs.

Intended Use

This model is engineered to serve as a core component in enterprise-grade Arabic/multilingual NLP pipelines, specifically:

Retrieval-Augmented Generation (RAG): Chunking unstructured documents into highly cohesive text segments prior to vector embedding generation.
Advanced Preprocessing: A smart, context-aware alternative to naive rule-based text splitters (e.g., regex, token count limits) which frequently break up semantic contexts in complex Arabic phrasing.
Technical Document Analysis: Flawlessly handling technical Arabic material heavily interspersed with English terminology.
Information Extraction (IE): Isolating independent factual claims or narrative transitions for downstream analytical tasks.

Caution: This model is strictly a sequence segmenter. It is not designed for abstractive generation, translation, open-ended QA, or text summarization.

Training Details

LoRA Configuration

Table with columns: Parameter, Value
Parameter	Value
LoRA Rank (r)	64
LoRA Alpha	64
LoRA Dropout	0
Target Modules	`q_proj`, `k_proj`, `v_proj`, `o_proj`, `gate_proj`, `up_proj`, `down_proj`
Bias

Supervised Fine-Tuning (SFT) Hyperparameters

Table with columns: Parameter, Value
Parameter	Value
Training Epochs	3
Per-Device Train Batch Size	1
Gradient Accumulation Steps	4
Effective Batch Size	4
Learning Rate	1e-4
LR Scheduler Type	Cosine
Warmup Ratio	0.1
Optimizer	`adamw_8bit`

The final adapter utilized in production corresponds to checkpoint-1000, chosen for its optimal evaluation convergence and stable validation loss profile without indicating signs of overfitting.

Dataset Architecture

The dataset was constructed via a rigorous Knowledge Distillation pipeline utilizing gemini-3.1-flash-lite as the Teacher Model to generate high-quality ground-truth mappings (Raw Text → Syntactically Valid JSON Chunks).

Data Sources & Distribution

Three distinct text styles were synthesized to ensure cross-domain generalization:

Table with columns: Source, Description, Count
Source	Description	Count
Wikipedia (Modern Standard Arabic)	Formal, encyclopedic prose	568
XLSum (BBC Arabic News)	Direct, objective journalistic style	1476
Technical Wikipedia (Mixed)	Code-switched documents (5%–40% English token density)	239
Total Corpus Size		2,257 Articles

Data Splits

Table with columns: Split, Sample Count
Split	Sample Count
Train	2000
Validation	257

Each training sequence wraps the original raw layout into an instruction text prompt, with a strictly deterministic, character-exact JSON schema mapping as the targeted output.

Infrastructure and Hardware

Table with columns: Component, Details
Component	Details
Graphics Processing Unit (GPU)	NVIDIA Tesla T4
Environment Environment	Google Colab (Free Tier)
Training Optimization Stack	Unsloth AI + Hugging Face TRL
Experiment Tracking	Weights & Biases (WandB)

Integrating Unsloth was vital to successfully execute this training pipeline on a commodity T4 GPU's restricted VRAM profile (15GB), providing nearly a 2x acceleration advantage alongside minimal hardware activation overhead.

Load Testing & Performance

Following convergence, the model was deployed via the high-concurrency vLLM engine and stress-tested using Locust to simulate realistic concurrent production workloads.

Serving Configuration (vLLM)

Table with columns: Metric / Parameter, Configured Value
Metric / Parameter	Configured Value
Inference Engine	vLLM Engine Process
GPU Memory Utilization	0.8 (80% of VRAM allocated to KV Cache)
Max Model Length (`max_model_len`)	2048 Tokens
Max LoRA Rank	64
Inference Precision	`half` (Float16 Execution)

Performance Benchmark Under Stress

A sustained load test simulating simultaneous clients generating continuous validation payloads yields the following verified metrics:

Table with columns: Performance Metric, Evaluation Result
Performance Metric	Evaluation Result
Inference Throughput	437.20 Tokens / Second
GPU KV Cache Usage:	~3.2%(Significant headroom remaining)
Request Failure Rate	0.00% (Zero Errors Over Duration)
Sustained Duration	60 Seconds
Concurrency Load Tool	Locust (Simulating 20 Concurrent Users)

Architectural Note: These benchmarks were successfully conducted under a concurrent load of 20 users. vLLM's Continuous Batching and PagedAttention maximized serving stability, keeping the GPU KV Cache usage at a highly efficient ~3.2% with zero out-of-memory (OOM) or timeout constraints.

Usage Guide

Loading via Unsloth (Recommended for Inference)

python

from unsloth import FastLanguageModel # Load the model and tokenizer natively using Unslothmodel, tokenizer = FastLanguageModel.from_pretrained(    model_name     = "Mo-Abdelfattah/arabic-semantic-chunker-qwen1.5b",    max_seq_length = 3500,    dtype          = None,    load_in_4bit   = False,)FastLanguageModel.for_inference(model)

Loading via Standard Hugging Face Transformers

python

from transformers import AutoModelForCausalLM, AutoTokenizerfrom peft import PeftModel # Load base model firstbase_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-1.5B-Instruct") # Wrap with the LoRA adapter weightsmodel = PeftModel.from_pretrained(base_model, "Mo-Abdelfattah/arabic-semantic-chunker-qwen1.5b")tokenizer = AutoTokenizer.from_pretrained("Mo-Abdelfattah/arabic-semantic-chunker-qwen1.5b")

Production Serving via vLLM CLI

bash

vllm serve "Qwen/Qwen2.5-1.5B-Instruct" \  --dtype=half \  --gpu-memory-utilization 0.8 \  --max-model-len 2048 \  --max-lora-rank 64 \  --enable-lora \  --lora-modules arabic-chunker="Mo-Abdelfattah/arabic-semantic-chunker-qwen1.5b"

Output Format

The model strictly adheres to outputting valid, parsable JSON matching the following structural schema:

json

{  "original_text_length": 2345,  "semantic_chunks": [    "First coherent semantic segment extracted from the text...",    "Second logical chunk containing continuing semantic context..."  ]}

Limitations & Boundary Conditions

Strict Task Bound: The model is exclusively trained for text sequence segmentation. Abstractive reasoning, creative generation, or open-domain synthesis tasks are explicitly unsupported.
Context Length Restrictions: The training pipeline sliced input sources at a maximum boundary of 1,500 characters per document sample; performance on exceedingly long inputs (e.g., >10k tokens) may degrade.
Distillation Dependency: The training data utilizes synthesized extractions from gemini-3.1-flash-lite, meaning that any localized logical parsing bias present in the teacher model may occasionally be inherited.

Author

Mohamed Abdelfattah AI & Data Platform Engineer specializing in high-throughput Arabic NLP architectures, robust LLM alignment pipelines, and optimal distributed serving infrastructures.

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Container

Run this model inference with full control and performance in your environment.

Learn more

Model Details

Model Provider

Mo-Abdelfattah

Model Tree

Base

Qwen/Qwen2.5-1.5B-Instruct

Adapter

this model

Input Modalities

Text

Output Modalities

Text

Supported Functionality

Dedicated EndpointsContainer

Explore FriendliAI today

Get started Talk to an engineer

README

License: apache-2.0

Model Overview
Intended Use
Training Details
Dataset Architecture
Infrastructure & Hardware
Load Testing & Performance
Usage Guide
Output Schema
Limitations & Boundary Conditions

Model Overview

Table with columns: Attribute, Value
Attribute	Value
Base Model	`Qwen/Qwen2.5-1.5B-Instruct`
Task	Text Segmentation / Multilingual Semantic Chunking
Finetuning Method	Supervised Fine-Tuning (SFT) with LoRA
Framework Stack	Unsloth + Hugging Face TRL (`SFTTrainer`)
Supported Languages	Arabic, English, and Code-Switched (Mixed) Text
License	Apache 2.0

Intended Use

This model is engineered to serve as a core component in enterprise-grade Arabic/multilingual NLP pipelines, specifically:

Retrieval-Augmented Generation (RAG): Chunking unstructured documents into highly cohesive text segments prior to vector embedding generation.
Advanced Preprocessing: A smart, context-aware alternative to naive rule-based text splitters (e.g., regex, token count limits) which frequently break up semantic contexts in complex Arabic phrasing.
Technical Document Analysis: Flawlessly handling technical Arabic material heavily interspersed with English terminology.
Information Extraction (IE): Isolating independent factual claims or narrative transitions for downstream analytical tasks.

Caution: This model is strictly a sequence segmenter. It is not designed for abstractive generation, translation, open-ended QA, or text summarization.

Training Details

LoRA Configuration

Table with columns: Parameter, Value
Parameter	Value
LoRA Rank (r)	64
LoRA Alpha	64
LoRA Dropout	0
Target Modules	`q_proj`, `k_proj`, `v_proj`, `o_proj`, `gate_proj`, `up_proj`, `down_proj`
Bias

Supervised Fine-Tuning (SFT) Hyperparameters

Table with columns: Parameter, Value
Parameter	Value
Training Epochs	3
Per-Device Train Batch Size	1
Gradient Accumulation Steps	4
Effective Batch Size	4
Learning Rate	1e-4
LR Scheduler Type	Cosine
Warmup Ratio	0.1
Optimizer	`adamw_8bit`

The final adapter utilized in production corresponds to checkpoint-1000, chosen for its optimal evaluation convergence and stable validation loss profile without indicating signs of overfitting.

Dataset Architecture

Data Sources & Distribution

Three distinct text styles were synthesized to ensure cross-domain generalization:

Table with columns: Source, Description, Count
Source	Description	Count
Wikipedia (Modern Standard Arabic)	Formal, encyclopedic prose	568
XLSum (BBC Arabic News)	Direct, objective journalistic style	1476
Technical Wikipedia (Mixed)	Code-switched documents (5%–40% English token density)	239
Total Corpus Size		2,257 Articles

Data Splits

Table with columns: Split, Sample Count
Split	Sample Count
Train	2000
Validation	257

Each training sequence wraps the original raw layout into an instruction text prompt, with a strictly deterministic, character-exact JSON schema mapping as the targeted output.

Infrastructure and Hardware

Table with columns: Component, Details
Component	Details
Graphics Processing Unit (GPU)	NVIDIA Tesla T4
Environment Environment	Google Colab (Free Tier)
Training Optimization Stack	Unsloth AI + Hugging Face TRL
Experiment Tracking	Weights & Biases (WandB)

Load Testing & Performance

Following convergence, the model was deployed via the high-concurrency vLLM engine and stress-tested using Locust to simulate realistic concurrent production workloads.

Serving Configuration (vLLM)

Table with columns: Metric / Parameter, Configured Value
Metric / Parameter	Configured Value
Inference Engine	vLLM Engine Process
GPU Memory Utilization	0.8 (80% of VRAM allocated to KV Cache)
Max Model Length (`max_model_len`)	2048 Tokens
Max LoRA Rank	64
Inference Precision	`half` (Float16 Execution)

Performance Benchmark Under Stress

A sustained load test simulating simultaneous clients generating continuous validation payloads yields the following verified metrics:

Table with columns: Performance Metric, Evaluation Result
Performance Metric	Evaluation Result
Inference Throughput	437.20 Tokens / Second
GPU KV Cache Usage:	~3.2%(Significant headroom remaining)
Request Failure Rate	0.00% (Zero Errors Over Duration)
Sustained Duration	60 Seconds
Concurrency Load Tool	Locust (Simulating 20 Concurrent Users)

Architectural Note: These benchmarks were successfully conducted under a concurrent load of 20 users. vLLM's Continuous Batching and PagedAttention maximized serving stability, keeping the GPU KV Cache usage at a highly efficient ~3.2% with zero out-of-memory (OOM) or timeout constraints.

Usage Guide

Loading via Unsloth (Recommended for Inference)

python

from unsloth import FastLanguageModel # Load the model and tokenizer natively using Unslothmodel, tokenizer = FastLanguageModel.from_pretrained(    model_name     = "Mo-Abdelfattah/arabic-semantic-chunker-qwen1.5b",    max_seq_length = 3500,    dtype          = None,    load_in_4bit   = False,)FastLanguageModel.for_inference(model)

Loading via Standard Hugging Face Transformers

python

from transformers import AutoModelForCausalLM, AutoTokenizerfrom peft import PeftModel # Load base model firstbase_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-1.5B-Instruct") # Wrap with the LoRA adapter weightsmodel = PeftModel.from_pretrained(base_model, "Mo-Abdelfattah/arabic-semantic-chunker-qwen1.5b")tokenizer = AutoTokenizer.from_pretrained("Mo-Abdelfattah/arabic-semantic-chunker-qwen1.5b")

Production Serving via vLLM CLI

bash

vllm serve "Qwen/Qwen2.5-1.5B-Instruct" \  --dtype=half \  --gpu-memory-utilization 0.8 \  --max-model-len 2048 \  --max-lora-rank 64 \  --enable-lora \  --lora-modules arabic-chunker="Mo-Abdelfattah/arabic-semantic-chunker-qwen1.5b"

Output Format

The model strictly adheres to outputting valid, parsable JSON matching the following structural schema:

json

{  "original_text_length": 2345,  "semantic_chunks": [    "First coherent semantic segment extracted from the text...",    "Second logical chunk containing continuing semantic context..."  ]}

Limitations & Boundary Conditions

Strict Task Bound: The model is exclusively trained for text sequence segmentation. Abstractive reasoning, creative generation, or open-domain synthesis tasks are explicitly unsupported.
Context Length Restrictions: The training pipeline sliced input sources at a maximum boundary of 1,500 characters per document sample; performance on exceedingly long inputs (e.g., >10k tokens) may degrade.
Distillation Dependency: The training data utilizes synthesized extractions from gemini-3.1-flash-lite, meaning that any localized logical parsing bias present in the teacher model may occasionally be inherited.

Author

Mohamed Abdelfattah AI & Data Platform Engineer specializing in high-throughput Arabic NLP architectures, robust LLM alignment pipelines, and optimal distributed serving infrastructures.

arabic-semantic-chunker-qwen1.5b

README

Table of Contents

Model Overview

Intended Use

Training Details

LoRA Configuration

Supervised Fine-Tuning (SFT) Hyperparameters

Dataset Architecture

Data Sources & Distribution

Data Splits

Infrastructure and Hardware

Load Testing & Performance

Serving Configuration (vLLM)

Performance Benchmark Under Stress

Usage Guide

Loading via Unsloth (Recommended for Inference)

Loading via Standard Hugging Face Transformers

Production Serving via vLLM CLI

Output Format

Limitations & Boundary Conditions

Author

Explore FriendliAI today

README

Table of Contents

Model Overview

Intended Use

Training Details

LoRA Configuration

Supervised Fine-Tuning (SFT) Hyperparameters

Dataset Architecture

Data Sources & Distribution

Data Splits

Infrastructure and Hardware

Load Testing & Performance

Serving Configuration (vLLM)

Performance Benchmark Under Stress

Usage Guide

Loading via Unsloth (Recommended for Inference)

Loading via Standard Hugging Face Transformers

Production Serving via vLLM CLI

Output Format

Limitations & Boundary Conditions

Author