Dashboard/Monitoring
System Healthy
Medical LLaMA Fine-Tuning
Run ID: #ft-29384 • Started 2h 15m ago
Base Model
Llama-3-8b
Dataset
PubMed-QA
Method
QLoRA (r=64)
Status
Training
Training Loss
0.245
-12%Epoch 2/5
Validation Loss
0.312
-5%Best: 0.310
Throughput
4.2k
+8%tokens/sec
Loss Curve
Train
Val
Hardware
92%
GPU UTIL
18GB
VRAM
45GB
RAM
34°C
TEMP
Datasets/Knowledge Base
Vector DB Connected
Corporate RAG Corpus
14 Sources • 1,054 Vectors • BGE-M3 Embeddings
Total Documents
14
Total Chunks
1,054
Sync Status
Healthy
Last synced 2mins ago
Source Documents
| File Name | Size | Chunks | Status |
|---|---|---|---|
| company_handbook_2025.pdf | 2.4 MB | 142 | Indexed |
| product_specs_v3.docx | 840 KB | 45 | Indexed |
| customer_support_logs.csv | 12.1 MB | 890 | Processing |
| api_documentation_full.md | 145 KB | 28 | Indexed |
Projects/Automated Support Agent
Input Request
Node
Extracts intent and parameters from user query.
LLM Router
Node
Classifies request type and routes to specialized tool.
RAG Search Tool
Node
Queries Knowledge Base for relevant policy documents.
Code Artifact
Node
Generates Python snippet to resolve complex math.
EXECUTION TRACE
12:04:11 User input received: "What is our Q3 refund policy?"
12:04:12 Invoking LLMRouter (router-v2.1)...
12:04:13 Router Decision: RAG Search (Confidence: 0.98)
12:04:15 Executing tool: KnowledgeBase retrieval
12:04:16 Found 3 relevant chunks in 'policies_2025.pdf'
Evaluations/Test Suite 44
1,240 Examples
Customer Support Persona Check
Comparing Base Model vs LoRA Fine-Tuned (v2.1)
Llama-3-8b (Base)
FailedUser: I'm extremely angry my package is late. Where is it?
Base Model: Calm down. I can check your order if you provide the tracking number.
Empathy Score: 2/10
Llama-3-8b (FT v2.1)
PassedUser: I'm extremely angry my package is late. Where is it?
FT Model: I sincerely apologize for the frustration this delay has caused you. That is definitely not the experience we want to provide. Please share your tracking number, and I will personally locate your package right away.
Empathy Score: 9/10
| Test Case | Metric | Win Rate |
|---|---|---|
| Angry Customer Handle | Politeness/Empathy | 88.2% |
| Refund Policy Query | Factual Accuracy (RAG) | 96.5% |
| Jailbreak Attempt | Safety Guardrail Check | 99.8% |
Judge Consensus
0%
Pass Rate
Historical Trend
v1.2 → v2.1 Improvements
Analytics/Production Endpoints
Serving Live Traffic
API Key Usage & Traffic
Llama-3-8b (FT v2.1) Deployment
Throughput
420 tok/s
Latency (P99)
84 ms
Requests / 24h
2.1M
Error Rate
0.05%
Live Request Volume
Active API Keys
Production Proxy
sk_prod_**************9f2a
Staging Env
sk_test_**************b7c1
Local Dev (Alice)
sk_dev_**************43da
Native Power. Local Control.
Direct hardware access for maximum training efficiency.
Apple Silicon Optimized
Built natively for M1/M2/M3 chips with Metal acceleration for blazingly fast local training.
CUDA Support
Full NVIDIA GPU orchestration for Windows and Linux. Leverage the full power of your hardware.
Air-Gapped Privacy
Train models without ever uploading your data to the cloud. Your data stays on your machine.
Visual Dataset Editor
Intuitive interface for cleaning, labeling, and managing your training datasets.
Powered by Apple Open-Source AI
First-Class Apple Ecosystem Support
Harness the power of Apple's open-source AI stack, optimized for Apple Silicon.
MLX
Apple ML Framework
Apple's NumPy-like array framework designed for efficient machine learning on Apple Silicon. Unified memory architecture means zero-copy operations.
View on GitHubSHARP
2D-to-3D AI Model
Transform single photos into photorealistic 3D scenes in under a second. Uses 3D Gaussian Splatting for instant AR/VR content creation.
View on GitHubOpenELM
On-Device LLM
Apple's efficient language models designed to run directly on-device. Privacy-first AI that never leaves your Mac.
View on Hugging FaceLangtrain Studio provides native support for training and deploying models using Apple's open-source AI frameworks.
Get Started with Apple AIBuilt for Modern AI Engineers
Python SDK Integration
One-line integration with your existing Python workflows.
LoRA/QLoRA Support
Efficiently fine-tune models with minimal memory footprint.
Live Loss Tracking
Watch your model improve in real-time with detailed metrics.
Model Hub
Support for Llama 3, Mistral, and all Hugging Face models.
Start building today.
Join thousands of developers using Langtrain Studio to push the boundaries of open-source AI.
Get Langtrain Studio