chore(misc): minor improvements

README.md

@@ -8,7 +8,7 @@ sdk_version: 5.19.0
 app_file: app.py
 pinned: false
 license: gpl-3.0
-short_description: Perplexity calculations
+short_description: PPL calculations
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -41,11 +41,10 @@ def create_interface() -> gr.Blocks:
     manager = ModelManager()
 
     with gr.Blocks() as demo:
-        gr.Markdown("# Language Model Perplexity Calculator (Sequential Version)")
-        gr.Markdown("Enter Hugging Face model checkpoint names (comma-separated).")
+        gr.Markdown("# LLM Perplexity")
 
         checkpoint_input = gr.Textbox(
-            label="Checkpoints (e.g. Qwen/Qwen2.5-14B-Instruct)",
+            label="Checkpoints",
             value=", ".join(checkpoints),
         )
 
@@ -53,7 +52,7 @@ def create_interface() -> gr.Blocks:
         perplexity_btn = gr.Button("Calculate Perplexity")
 
         load_output = gr.Textbox(label="Model Loading Status", interactive=False)
-        perplexity_output = gr.JSON(label="Perplexity Results")
+        perplexity_output = gr.JSON(label="PPL Results")
 
         # Connect event handlers
         load_btn.click(

utils.py

@@ -43,8 +43,8 @@ def _perplexity(model, tokenizer, text):
     return torch.exp(torch.tensor(loss)).item()
 
 
-num_samples: Final[int] = 1500  # Sample size for perplexity calculation
-sample_length: Final[int] = 128  # Maximum length of text to consider for perplexity
+num_samples: Final[int] = 500  # Sample size for perplexity calculation
+sample_length: Final[int] = 100  # Maximum length of text to consider for perplexity
 
 loaded = load_model(checkpoints)