Research · Open Weights

LumiChats Labs

Open-source fine-tuned language models for conversational AI, function calling, vision, medical imaging, and speech synthesis. All freely available on Hugging Face.

7
Open Models
7
On Hugging Face
5
Base Architectures
3
Open Licenses
Conversational

LumiChats v1.1

v1.1

Efficient multilingual conversational AI on Llama 3.2 3B

Llama-3.2-3B-InstructJun 2025Llama
3,237,063,680 parameters
  • ●Only 0.75% of parameters trained via LoRA — 24.3M of 3.21B
  • ●2x faster inference with Unsloth optimizations
  • ●2.35 GB peak additional memory during training on Tesla T4
Hugging FaceView Research
Function Calling

LumiChat Coder v2.1

v2.1

Specialised JSON function-calling and tool-use AI for production agents

Qwen2.5-Coder-1.5B-InstructAug 2025Apache
1,540,000,000 parameters
  • ●96.5% correct function/tool selection accuracy
  • ●99.8% JSON validity rate with grammar-constrained decoding (transformers-CFG)
  • ●94.2% properly formatted argument generation
Hugging FaceView Research
Vision / OCR

LumiChats v1.2 7B

v1.2

Vision-language model for converting handwritten math to LaTeX

Qwen2.5-VL-7B-Instruct-bnb-4bitSep 2025Apache
~8,343,688,192 parameters
  • ●Trained on 68,686 handwritten formula image-LaTeX pairs
  • ●Only 51.5M of 8.3B parameters trained (0.62%) via LoRA
  • ●Training completed in 3.27 minutes on Tesla T4
Hugging FaceView Research
Medical Vision

LumiChats v1.3 11B Vision

v1.3

Radiology-specialised vision model for clinical image description

Llama-3.2-11B-Vision-Instruct-bnb-4bitDec 2024Llama
~11B parameters
  • ●Substantially reduced hallucination rate vs. Llama 3.2 11B Vision Instruct
  • ●Trained on expert-annotated radiological image captions
  • ●60% memory reduction via 4-bit quantisation (Unsloth + bitsandbytes)
Hugging FaceView Research
Conversational

LumiChats 4B v1.4

v1.4

Gemma-3-4B fine-tuned for natural multi-turn dialogue

gemma-3-4b-it (Google DeepMind)Oct 2025Gemma
4,314,980,720 parameters
  • ●Only 0.35% of parameters trained — 14.9M of 4.31B
  • ●+15% conversational coherence over Gemma-3-4B-IT (estimated)
  • ●+20% multi-turn context retention (estimated)
Hugging FaceView Research
Text-to-Speech

LumiChats TTS 1B

v1.0

Voice-consistent text-to-speech model with reference audio cloning

csm-1b (Context-Speech-Model)Jul 2025Apache
~1 parameters
  • ●Only 1.75% of parameters updated via LoRA — efficient speaker adaptation
  • ●Reference audio conditioning for voice cloning
  • ●Maintains speaker identity across multi-sentence generation
Hugging FaceView Research
Instruction

LumiChats Instruct 14B

v1.0

14B parameter Phi-4 fine-tuned for structured multi-turn conversations

phi-4Jan 2026MIT
14,725,043,200 parameters
  • ●Built on Phi-4 14B — outperforms GPT-4o on GPQA graduate science (56.1% vs 50.6%)
  • ●80.4% on MATH competition problems; 84.8% on MMLU general knowledge
  • ●Only 65.5M of 14.7B parameters updated (0.45%) via LoRA
Hugging FaceView Research
All models free & open source

Use these models in LumiChats

Access LumiChats Labs models alongside 40+ frontier AI — Claude, GPT, Gemini — all in one platform.

Start for free →View all on Hugging Face
All models released by Aditya Kumar Jha · LumiChats Labs · Free for research and commercial use under their respective licenses