Upload README.md with huggingface_hub

README.md

@@ -1,23 +1,19 @@
-# nanochat-eos (EOS-chat)
+# nanochat EOS-chat (SFT)
 
-Energy LLM **EOS-chat** model and tokenizer, developed at Safire by the AI team.
+Nanochat SFT checkpoint and tokenizer for EOS-chat.
 
-## Layout
+## Files
 
-- **tokenizer/** – BPE tokenizer (`tokenizer.pkl`, `token_bytes.pt`). Used by all checkpoints.
-- **base_checkpoints/d26/** – Pretrained base model (step 7226): `model_007226.pt`, `meta_007226.json`.
-- **chatsft_checkpoints/d26/** – SFT (chat) model (step 848): `model_000848.pt`, `meta_000848.json`.
+- `chatsft_checkpoints/d26/model_000704.pt` – model state dict
+- `chatsft_checkpoints/d26/meta_000704.json` – config and metadata
+- `tokenizer/` – tokenizer files
 
-## Loading locally
+## Load with nanochat
 
-Set `NANOCHAT_BASE_DIR` to the directory that contains `tokenizer/`, `base_checkpoints/`, and `chatsft_checkpoints/` (e.g. this repo root after cloning), then:
+Clone this repo and set it as the nanochat base directory:
 
 ```bash
-# Chat (SFT) model
-python -m scripts.chat_cli
-
-# Or specify phase and tag
-python -m scripts.chat_cli --model-tag d26 --phase sft
+git clone https://huggingface.co/ksjpswaroop/nanochat-eos nanochat-eos
+export NANOCHAT_BASE_DIR=$(pwd)/nanochat-eos
+python -m scripts.chat_cli  # use SFT model
 ```
-
-The scripts resolve `get_base_dir()` to `NANOCHAT_BASE_DIR` and load the tokenizer and latest checkpoint from `chatsft_checkpoints/d26` (or `base_checkpoints/d26` for base).