Research at the speed of thought.

Qualia is the research agent that ideates, runs experiments, and collaborates with you.

~/lm-training › train_transformer.ipynb

[1]

from lm_lab import TransformerLM, Trainer
from lm_lab.data import WikiText103
import matplotlib.pyplot as plt

Executed in 611ms

[2]

model = TransformerLM(
    n_layers=12, n_heads=12, d_model=768,
    max_seq_len=1024, dropout=0.1,
).cuda()
print(f"Parameters: {model.num_parameters():,}")

Parameters: 124,439,808

Executed in 610ms

[3]

trainer = Trainer(
    model,
    dataset=WikiText103(),
    lr=6e-4, weight_decay=0.1,
    scheduler="cosine", min_lr=6e-5,
)

Executed in 609ms

Set up a training loop for a 125M parameter transformer on WikiText-103 with AdamW and cosine annealing

I’ll wire up the full training pipeline: load/tokenize WikiText-103, build a ~125M decoder-only Transformer, and set up AdamW plus a cosine learning-rate schedule. I’ll also add common Transformer training tricks (mixed precision, gradient clipping, checkpointing, periodic validation perplexity) so the run is stable and results are comparable.

Made 3 changes

Setup is done. I’ll treat validation perplexity as the headline metric, evaluate on a fixed interval, and checkpoint so we can do apples-to-apples comparisons when we start testing improvements.

GPT-5.2 (Low) ›

Qualia is a love letter to the art of research.

Built by researchers to make practical research workflows faster and easier.

From idea to results.

Qualia runs wherever Jupyter does. Chat with it to clean data, make beautiful visualizations, and fit models—just by asking.

Explore every direction.

Never leave an idea on the table again. Qualia coordinates parallel agents so you can iterate on promising model variants and re-examine that featurization choice you made last week.

The best research assistant.

Propose a direction in Slack, and Qualia runs for hours—training models, testing variations, and reporting what works.

“This feels like a superpower.”

Lino Le VanCo-founder at AlphaXiv

Pricing

Qualia Free

Explore Qualia

$0for ever

Chat-based data analysis
Community support

Pro

For researchers

$20Per month

Everything in Free, plus:

Up to 8 parallel agents: explore multiple directions at once
Access to GPU nodes
Priority support

Max

For power users

$100Per person billed monthly

Everything in Pro, plus:

Autonomous multi-hour research runs
Slack / Google Docs integration
Higher output limits for all tasks
Dedicated support

Enterprise

For organizations with custom needs

Custom

On-prem deployment—your IP stays local
Integration with your codebase and internal tools
Dedicated field engineering
24/7 support with SLAs
SSO, audit logs, and compliance controls