It is trained with PoSTMEDIA's in-house Capability-Preserving Full Fine-Tuning recipe — a full-parameter SFT pipeline designed so that customization does not erode the reasoning, instruction-following, and multilingual abilities of the Gemma-4 base model.

Compared to Lux-V1, Lux-V1-Pro adapts a larger, dense 31B base with all parameters trainable, targeting maximum capability for demanding downstream tasks.

Highlights

Full-parameter fine-tuning of Gemma-4-31B (dense) — every weight is updated
Base capability preserved — pretraining knowledge and reasoning skills remain intact after SFT
Dataset-flexible — any combination of curated instruction / domain / persona datasets can be composed into a single full-FT run
Maximum capability tier of the Lux line, intended for the most demanding reasoning and generation workloads

Model Overview

Specification	Details
Base Model	`google/gemma-4-31B-it`
Parameters	31B (dense)
Architecture	Decoder-only Transformer (dense)
Training Precision	BF16
Inference Precision	BF16
Context Length	Inherits from Gemma-4 base
Fine-Tuning Method	Full-parameter SFT (Capability-Preserving recipe)

Capability-Preserving Full Fine-Tuning

Naive full fine-tuning of large pretrained LLMs often damages the base model's general abilities — a well-known trade-off when SFT is pushed too far. PoSTMEDIA's recipe is built specifically to avoid this.

For Lux-V1-Pro, three design choices keep the Gemma-4 base intact while still allowing deep adaptation:

All parameters trainable, conservatively. As a dense model, Lux-V1-Pro updates every weight — but under a tightly controlled optimization regime that keeps the model in the neighborhood of the pretrained distribution.
Architecture-tuned learning rate. A lower LR is used for the 31B dense backbone, deliberately calibrated to avoid the catastrophic-forgetting regime that aggressive full-FT typically falls into.
Continuous base-capability evaluation. Evaluation runs at the start of training and at every epoch, so any regression in base-model quality is caught early rather than discovered post-hoc.

This means Lux-V1-Pro can be re-trained from the same base with arbitrary mixtures of datasets — identity, domain knowledge, instruction-style, reasoning — without losing what Gemma-4 already knows.

Training Configuration

Parameter	Value
Fine-Tuning Method	Full-parameter SFT (all weights trainable)
Precision	BF16
Distributed Strategy	DeepSpeed ZeRO-3 + CPU offload
Training Infrastructure	NVIDIA H200 × 8

Quick Start

pip install transformers accelerate

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_name = "PoSTMEDIA/Lux-V1-Pro"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

prompt = "Explain why preserving base-model capability matters during fine-tuning."
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Use Cases

High-capability enterprise assistants and reasoning agents
Domain-specialized models that must retain strong general-purpose abilities
Persona / identity-aligned chat with deep instruction following
Downstream tasks where the larger dense backbone outperforms the MoE tier

Safety & Limitations

Inherits the safety characteristics of the Gemma-4 base; output guardrails are recommended for production.
Not intended for medical, legal, or financial decision-making.
May occasionally hallucinate — human review is recommended for critical outputs.

Citation

bibtex

@misc{lux_v1_pro_2026,
  title  = {Lux-V1-Pro: Capability-Preserving Full Fine-Tuning of Gemma-4-31B},
  author = {PoSTMEDIA AI Lab},
  year   = {2026},
  publisher = {Hugging Face}
}

Contact

PoSTMEDIA AI Lab