Consolidate to single tokenizer and fix CSS corner clipping

- Removed dual tokenizer setup (MistralTokenizer + AutoTokenizer)
- Now using single AutoTokenizer with tokenizer_type="mistral"
- Use apply_chat_template() instead of encode_chat_completion()
- Added overflow:visible CSS to prevent rounded corner clipping

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

app.py

@@ -17,8 +17,6 @@ import spacy
 import spacy.cli
 import torch
 from mistralai import Mistral
-from mistral_common.protocol.instruct.request import ChatCompletionRequest
-from mistral_common.tokens.tokenizers.mistral import MistralTokenizer
 from transformers import AutoTokenizer, Mistral3ForConditionalGeneration, TextIteratorStreamer
 
 
@@ -27,7 +25,6 @@ from transformers import AutoTokenizer, Mistral3ForConditionalGeneration, TextIt
 # ============================================================================
 
 MODEL_ID = "ellamind/sui-1-24b"
-HF_TOKENIZER_ID = "mistralai/Mistral-Small-3.1-24B-Instruct-2503"
 
 LANGUAGES = {
     "English": "en",
@@ -57,24 +54,31 @@ def download_spacy_models():
 
 
 def load_model():
-    """Load the sui-1-24b model and tokenizers."""
+    """Load the sui-1-24b model and tokenizer."""
     print(f"Loading model: {MODEL_ID}")
-    
-    tokenizer = MistralTokenizer.from_hf_hub(MODEL_ID)
+
     model = Mistral3ForConditionalGeneration.from_pretrained(
         MODEL_ID,
         dtype=torch.bfloat16,
         device_map="auto",
     )
-    hf_tokenizer = AutoTokenizer.from_pretrained(HF_TOKENIZER_ID)
-    
+
+    tokenizer = AutoTokenizer.from_pretrained(
+        MODEL_ID,
+        trust_remote_code=True,
+        tokenizer_type="mistral",
+        use_fast=False,
+    )
+    if tokenizer.pad_token is None:
+        tokenizer.pad_token = tokenizer.eos_token
+
     print("Model loaded successfully")
-    return model, tokenizer, hf_tokenizer
+    return model, tokenizer
 
 
 # Initialize on startup
 download_spacy_models()
-model, tokenizer, hf_tokenizer = load_model()
+model, tokenizer = load_model()
 nlp_cache: dict[str, spacy.Language] = {}
 
 
@@ -319,20 +323,23 @@ def generate_summary(
     """Generate summary with streaming output."""
     tagged_text, tag_to_sentence, tag_to_pos = tag_sentences(text, lang_code)
     num_sentences = len(tag_to_sentence)
-    
+
     prompt = build_prompt(tagged_text, words, language_name, custom_instruction)
     messages = [{"role": "user", "content": prompt}]
-    tokenized = tokenizer.encode_chat_completion(ChatCompletionRequest(messages=messages))
-    inputs = torch.tensor([tokenized.tokens]).to(model.device)
+    inputs = tokenizer.apply_chat_template(
+        messages,
+        return_tensors="pt",
+        add_generation_prompt=True,
+    ).to(model.device)
 
-    streamer = TextIteratorStreamer(hf_tokenizer, skip_prompt=True, skip_special_tokens=True)
+    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
 
     gen_kwargs = {
         "input_ids": inputs,
         "streamer": streamer,
         "max_new_tokens": 4096,
         "do_sample": False,
-        "pad_token_id": hf_tokenizer.eos_token_id,
+        "pad_token_id": tokenizer.pad_token_id,
     }
     
     thread = Thread(target=model.generate, kwargs=gen_kwargs)
@@ -455,6 +462,15 @@ CSS = """
     max-width: 1200px !important;
     margin: 0 auto !important;
     padding: 1.5rem 2rem !important;
+    overflow: visible !important;
+}
+
+/* Prevent parent elements from clipping corners */
+.gradio-container > div,
+.gradio-row,
+.contain,
+.gap {
+    overflow: visible !important;
 }
 
 /* Header styling */
@@ -487,6 +503,7 @@ CSS = """
     border-radius: 0.75rem !important;
     border: 1px solid #e2e8f0 !important;
     padding: 1rem !important;
+    overflow: visible !important;
 }
 
 .output-panel {
@@ -495,6 +512,7 @@ CSS = """
     border: 1px solid #e2e8f0 !important;
     padding: 1rem !important;
     min-height: 300px;
+    overflow: visible !important;
 }
 
 /* Section headers - simpler, not uppercase */