
Bangla chatbot
বাংলা চ্যাটবটYour own ChatGPT-style assistant that answers in Bangla (Qwen2.5-7B).
Open this notebook ↗Read the short write-up →Bangla AI needs the same accelerator fundamentals as global AI workloads, but local datasets, speech, OCR and language-model research create their own practical constraints. ComputeBD is designed around those local workflows.
Shared by the ComputeBD team. Open one, change it to fit your work, then run it on an AI Computer — billed by the minute.

Your own ChatGPT-style assistant that answers in Bangla (Qwen2.5-7B).
Open this notebook ↗Read the short write-up →Fine-tune Whisper on FLEURS Bengali, then measure word error per region (Sylheti, Chittagong, Noakhali…).
Open this notebook ↗Read the short write-up →Train a text classifier on the BD-Dialect dataset; compare standard Bangla BERT models and their licences.
Open this notebook ↗Read the short write-up →Turn Bangla text into natural speech with described voices — with a clear consent-first voice-cloning policy.
Open this notebook ↗Read the short write-up →Build a Bangla sentiment dataset from commentary or talk-show transcripts and chart the mood minute by minute.
Open this notebook ↗Read the short write-up →Search transcripts by meaning and get Bangla answers that cite the minute they were said.
Open this notebook ↗Read the short write-up →Embed subtitles of educational videos or dramas and search by meaning, with links to the exact second.
Open this notebook ↗Read the short write-up →
Large model training, transformer workloads, HPC

LLM fine-tuning, research, high-throughput training

Generative AI, rendering, larger fine-tunes

Fast inference, video AI, efficient fine-tuning
Tell us your workload and preferred AI Computer.
Join waitlist ↗Memory (VRAM) needed for Llama, Qwen, Mistral, Stable Diffusion, FLUX… — which AI Computer fits, and the cost per hour in taka.
→Fine-tuning cost estimatorModel size, data and epochs → hours and cost in ৳. Print a one-page estimate for your supervisor or lab.
→Bangla token counterPaste Bangla text: see how many tokens it uses compared with English, and what that does to AI API bills.
→