Update README.md

README.md

@@ -41,14 +41,6 @@ with torch.inference_mode():
 print(tokenizer.decode(output[0], skip_special_tokens=True))
 ```
 
-Keep the weights in FP32 as shown: the implementation uses BF16 residuals and
-BF16 autocast internally, matching the training model's mixed-precision policy.
-The release retains the original FP32 master weights without lossy conversion.
-The tested path uses a single CUDA device. 
-Multi-device dispatch and quantization have not been validated. 
-Total prompt plus generation length must not exceed 4096.
-For batches, the tokenizer uses left padding; pass its `attention_mask` to the model.
-
 ## Architecture
 
 | Property | Value |