Thilak Ruwan Chamara Denipitiya

ruwan

2 2 106

ThilakDen

AI & ML interests

Recent Activity

liked a model about 23 hours ago

Jackrong/Qwopus3.6-27B-Fusion-GGUF

reacted to Enderchef's post with 🤗 1 day ago

Introducing GCI-Bench, Glint Research's first benchmark! This 5000-question bench benchmarks attention and gradients; it gives a prompt and unrelated filler parts. The goal is to benchmark if the model understands and attends to the important parts more than the unimportant parts. I've already benchmarked on the top 5 models on our Leaderboard(Glint2, Supra 50M, Glint1.3, GPT S 5M, GPT X2 125M) Pending benchmark status at https://huggingface.co/spaces/OpenEvals/README/discussions/2 https://huggingface.co/datasets/Glint-Research/GCI_Bench

reacted to FredyRivera-dev's post with 🚀 2 days ago

We wrote a full technical guide on how to train a bilingual (ES/EN) LLM from scratch: TinyQwen. Covers: - Hybrid architecture based on Qwen3.5 - Pre-training with 15B tokens - Cost benchmark between H200 and B200 - Post-training with SFT + LoRA - Full code and data, open source With ~$11 of compute on an H200 we ran an initial training run, enough to validate the full architecture and pipeline. Blog post: https://aquiles-ai.vercel.app/blog/tinyqwen-from-scratch Technical feedback welcome, especially from anyone looking to replicate the pipeline with more compute.

View all activity

Organizations

None yet

liked a model about 23 hours ago

Jackrong/Qwopus3.6-27B-Fusion-GGUF

Text Generation • 27B • Updated about 12 hours ago • 706 • 22

reacted to Enderchef's post with 🤗 1 day ago

Post

3715

Introducing GCI-Bench, Glint Research's first benchmark!

This 5000-question bench benchmarks attention and gradients; it gives a prompt and unrelated filler parts. The goal is to benchmark if the model understands and attends to the important parts more than the unimportant parts.

I've already benchmarked on the top 5 models on our Leaderboard(Glint2, Supra 50M, Glint1.3, GPT S 5M, GPT X2 125M)
Pending benchmark status at OpenEvals/README#2

Glint-Research/GCI_Bench

4 replies

reacted to FredyRivera-dev's post with 🚀 2 days ago

Post

6165

We wrote a full technical guide on how to train a bilingual (ES/EN) LLM from scratch: TinyQwen.

Covers:
- Hybrid architecture based on Qwen3.5
- Pre-training with 15B tokens
- Cost benchmark between H200 and B200
- Post-training with SFT + LoRA
- Full code and data, open source

With ~$11 of compute on an H200 we ran an initial training run, enough to validate the full architecture and pipeline.

Blog post: https://aquiles-ai.vercel.app/blog/tinyqwen-from-scratch

Technical feedback welcome, especially from anyone looking to replicate the pipeline with more compute.