Upload README.md with huggingface_hub

README.md

@@ -1,3 +1,19 @@
+---
+language:
+- en
+- ta
+license: mit
+tags:
+- diffusion
+- language-model
+- pytorch
+- shakespeare
+- tamil
+- nlp
+- from-scratch
+library_name: pytorch
+---
+
 # tiny-dllm
 
 A Masked Diffusion Language Model built from scratch in PyTorch — for learning and robotics research.
@@ -130,17 +146,6 @@ Logits [B, T, vocab_size]
 
 Both trained from scratch on TinyShakespeare with identical model size and steps.
 
-## Trained Checkpoint
-
-A checkpoint trained to 50,000 steps on TinyShakespeare is available on HuggingFace:
-
-```bash
-from huggingface_hub import hf_hub_download
-path = hf_hub_download(repo_id="sutharsan311/tiny-dllm", filename="dllm_step50000.pt")
-```
-
-Hardware: NVIDIA RTX 3050 (4GB VRAM) — ~2 hours training time.
-
 ## Roadmap
 
 - [x] Character-level tokenizer
@@ -150,7 +155,7 @@ Hardware: NVIDIA RTX 3050 (4GB VRAM) — ~2 hours training time.
 - [x] Tamil Unicode tokenizer
 - [x] Train on Thirukkural + Sangam poetry
 - [x] Tamil Wikipedia downloader (API + full dump)
-- [x] Train to 50k steps
+- [ ] Train to 50k steps
 - [ ] GPT baseline comparison (dLLM vs GPT on same data)
 - [ ] Blog post: dLLM vs GPT on TinyShakespeare
 - [ ] Fill-in-the-blanks (conditional generation)