Update app.py

app.py

@@ -136,24 +136,19 @@ def run_inference(message, history, model_picked, context_size, max_output):
     print(result)
     return result
 
-description="""
-A demo chat interface with Pixtral 12B EXL2 Quants, deployed using **ExllamaV2**!
-
+description="""A demo chat interface with Pixtral 12B EXL2 Quants, deployed using **ExllamaV2**!
 The model will be loaded once the GPU is available. This space specifically will load by default Pixtral at 4bpw from the following repository: [turboderp/pixtral-12b-exl2](https://huggingface.co/turboderp/pixtral-12b-exl2). Other quantization options are available.
-
 The current version of ExllamaV2 running is the dev branch, not the master branch: [ExllamaV2](https://github.com/turboderp/exllamav2/tree/dev).
 
-The model at **4bpw and 16k context size fits in less than 12GB of VRAM**!
+The model at **4bpw and 16k context size fits in less than 12GB of VRAM**, and at **2.5bpw and short context can potentially fit in 8GB of VRAM**!
 
 The current default settings are:  
 - Model Quant: 4.0bpw
 - Context Size: 16k tokens
 - Max Output: 512 tokens
-
 You can select other quants and experiment!
 
-Thanks, turboderp!
-"""
+Thanks, turboderp!"""
 examples = [
     [
         {"text": "What are the similarities and differences between these two experiments?", "files":["test_image_1.jpg", "test_image_2.jpg"]},