Fix card: param count 30m → 36.5m (untied lm_head omitted), weights fp16 → fp32 (#1)

- Fix card: param count 30m → 36.5m (untied lm_head omitted), weights fp16 → fp32 (ca22aa45fd5ecf4ba5a1b24e7d29e1c97a57cc3e)


Co-authored-by: Compactbot <Compactbot@users.noreply.huggingface.co>

README.md

@@ -13,8 +13,8 @@ tags:
 
 A small GPT model for code generation, trained completely from scratch (custom PyTorch architecture, no `transformers` library involved).
 
-- **Parameters**: 30m
-- **Weights format**: fp16
+- **Parameters**: 36.5m (36,505,728 learnable; the "30m" figure omitted the untied output head)
+- **Weights format**: fp32
 - **Architecture**: Decoder-only Transformer (custom implementation)
 - **Training**: from scratch, no pretrained weights used
 - **Language**: writes Python code, all comments and output in English