Updated GitHub markdown alerts (Lightning-AI#841)

tutorials/finetune_lora.md

@@ -167,17 +167,12 @@ python scripts/merge_lora.py \
   --out_dir "out/lora_merged/stablelm-base-alpha-3b/"
 ```
 
-> [!Note]
+> [!IMPORTANT]
 > If you changed the LoRA hyperparameters (`lora_r`, `lora_key`, etc.) in the
 > `finetune/lora.py` script, it is important to update the hyperparameter configuration
 > in the `scripts/merge_lora.py` script accordingly. Otherwise, you will encounter size
 > mismatch errors upon merging.
 
-> [!Note]
-> If the model was trained with quantized weights, it is possible to merge LoRA
-> and quantized pretrained weights by providing `--quantize` argument,
-> the same that was used for the finetuning stage.
-
 After merging, we can use the `base.py` file for inference using the new checkpoint file. Note that if your new checkpoint directory is different from the original checkpoint directory, we also have to copy over the tokenizer and config files:
 
 ```bash

tutorials/quantize.md

@@ -32,14 +32,14 @@ Memory used: 14.50 GB
 
 To reduce the memory requirements further, Lit-GPT supports several quantization techniques, which are shown below.
 
-> [!NOTE]
+> [!TIP]
 > Most quantization examples below also use the `--precision bf16-true` setting explained above. If your GPU does not support `bfloat16`, you can change it to `--precision 16-true`.
 
 ## `bnb.nf4`
 
 Enabled with [bitsandbytes](https://github.com/TimDettmers/bitsandbytes). Check out the [paper](https://arxiv.org/abs/2305.14314v1) to learn more about how it works.
 
-> [!NOTE]
+> [!IMPORTANT]
 > `bitsandbytes` only supports `CUDA` devices and the `Linux` operating system.
 > Windows users should use [WSL2](https://learn.microsoft.com/en-us/windows/ai/directml/gpu-cuda-in-wsl).