MokingbirdNode Hub

AI Development Platform — All systems operational

CPU Usage

34%

RAM Usage

6.2 GB

GPU VRAM

10.8 GB

Vector DB Size

2.4 GB

Modules

🦜

MokingbirdRAG

Production-Ready RAG Framework

24

Documents

1,847

Chunks

98%

Accuracy

ChromaDBOllamaMiniLMAdvanced

🧬

MbDataGen

AI Training Data Generation

8

Datasets

42K

Samples

5

Jobs

QA PairsInstructionsSynthetic

🎯

MbFineTuning

Model Fine-Tuning Management

3

Running

12

Completed

0.89

Best Loss

LoRAQLoRAPEFTLlama3

Recent Activity

RAG query completed — telecom_docs — "What is eCPRI specification?"

2m ago

Fine-tuning job llama3_telecom_v3 epoch 12/20 — loss: 0.234

5m ago

Dataset telecom_qa_v2 generated — 5,200 QA pairs completed

18m ago

Document uploaded — OAM_Guide_23.3.pdf — 847 chunks indexed

34m ago

System — Ollama service restarted, all models available

1h ago

🦜 MokingbirdRAG Advanced Pipeline

Pipeline Flow

Memory

Enhance

Retrieve

Format

Prompt

LLM

Analytics

Store

📁 Document Collection 24 docs

⬆️

Drop files here or click to upload

PDF, DOCX, TXT, CSV, JSON, MD, XLSX, PPTX, Images

📄 OAM_Guide_23.3.pdf 4.2MB
📄 eCPRI_spec_v2.1.pdf 2.8MB
📝 network_architecture.docx 1.1MB
📊 kpi_metrics.xlsx 520KB
📋 3gpp_standards.md 340KB
🖼️ architecture_diagram.png 1.8MB
📄 deployment_guide.pdf 3.1MB

⚙️ RAG Configuration

Pipeline Level

Embedding Model

LLM Provider

Retriever Strategy

Vector Store

Context Enhancement

Top-K Chunks

💬 Query Interface ● Ollama online

● llama3.2 via Ollama ⏱ 1.24s

The eCPRI (Enhanced Common Public Radio Interface) specification defines the interface between eREC (Radio Equipment Controller) and eRE (Radio Equipment) in fronthaul networks. In 5G deployments, eCPRI is critical for connecting Distributed Units (DUs) to Remote Radio Units (RRUs) over Ethernet-based fronthaul.

Key features of eCPRI v2.0 include:
• User Plane data transport over Ethernet/IP
• C&M Plane signaling for synchronization
• Support for time-sensitive networking (TSN)
• Maximum latency requirements: ≤100μs one-way

eCPRI_spec_v2.1.pdf §3.2 OAM_Guide_23.3.pdf §7.4 network_architecture.docx §2

Previous Query

"What are the KPI thresholds for 5G NR?"

The KPI thresholds for 5G NR as per 3GPP TS 28.552 include: DL throughput ≥100 Mbps for eMBB, latency ≤1ms for URLLC, and reliability ≥99.999% for critical applications...

📊 RAG Analytics Dashboard

Queries per Day (Last 7 Days)

Mon

Tue

Wed

Thu

Fri

Sat

Sun

Response Latency (ms) Trend

Retriever Strategy Distribution

Ensemble48%

MMR28%

Similarity15%

Multi-Query9%

Quality Metrics

Relevance Score0.87

Faithfulness0.92

Context Precision0.81

Answer Completeness0.78

Avg Latency1.24s

⚙️ RAG Configuration

Embedding Settings

Providersentence-transformers (local)

Modelall-MiniLM-L6-v2

Dimension384

Batch Size32

Normalize

Cache Embeddings

Vector Store Settings

BackendChromaDB

Path./rag_db/chroma

Collectiontelecom_docs

Distance Metriccosine

Persist

LLM Settings

ProviderOllama

Modelllama3.2:latest

Temperature0.7

Max Tokens2048

Streaming

Timeout60s

Chunking / Splitter Settings

StrategySemantic + Paragraph

Chunk Size512 tokens

Overlap50 tokens

Min Chunk100 tokens

Dedup

🧬 MbDataGen AI Data Generation

🎲 Generation Configuration

Generation Type

Source Documents

Generator LLM

Target Samples

Quality Filter Threshold

Output Format

Language

📊 Active Generation Jobs

telecom_qa_v3 Running

QA Pairs · gpt-4o · 5,200/10,000 samples

52% · ETA: 14 min · Quality: 0.91

legal_instructions_v1 Queued

Instruction Data · llama3.2 · 0/3,000 samples

🗄️ Completed Datasets

telecom_qa_v2JSONL

8,450 samples · 4.2MB · Quality: 0.89

medical_cot_v1JSONL

3,200 samples · 2.1MB · Quality: 0.94

🎯 MbFineTuning 3 Active Runs

llama3_telecom_v3● Running

0.234

Train Loss

0.271

Val Loss

12/20

Epoch

Method: QLoRA · Base: LLaMA-3-8B · Dataset: telecom_qa_v2

60% · ETA: 2h 14m · GPU: RTX 4090 · VRAM: 10.8/16GB

mistral_legal_lora⏳ Queued

Method: LoRA · Base: Mistral-7B · Dataset: legal_instructions_v1

Position 2 in queue · Starts after llama3_telecom_v3

⚙️ New Training Configuration

Base Model

Fine-tuning Method

Training Dataset

LoRA Rank (r)

LoRA Alpha

Learning Rate

Epochs

Batch Size

✅ Completed Models

llama3_telecom_v2GGUF + LoRA

Val Loss: 0.198 · 3h 22m training · 94.2% eval accuracy

⚙️ Global Settings

API Keys

OpenAI API Key

Cohere API Key

Pinecone API Key

Service Endpoints

Ollama Base URL

vLLM Endpoint

TGI Endpoint