| import gradio as gr |
| import torch |
| import json |
| import html |
| import traceback |
| from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
| print("Loading model...") |
| model_name = "Qwen/Qwen2.5-0.5B-Instruct" |
| tokenizer = AutoTokenizer.from_pretrained(model_name) |
| model = AutoModelForCausalLM.from_pretrained( |
| model_name, |
| torch_dtype=torch.float16, |
| device_map="auto" |
| ) |
| print("Model loaded.") |
|
|
| SYSTEM_PROMPT = """You are an English learning assistant. Extract 8-20 useful expressions from the text. |
| For each expression, output a JSON object with keys: expression, meaning, explanation, original_context, extra_example. |
| Meaning and explanation should be in Chinese. |
| Output must be a JSON array. No extra text.""" |
|
|
| def analyze(text): |
| try: |
| if not text or len(text.strip()) < 20: |
| return "<div style='color:red'>⚠️ Please enter at least 20 characters.</div>" |
|
|
| messages = [ |
| {"role": "system", "content": SYSTEM_PROMPT}, |
| {"role": "user", "content": text} |
| ] |
| inputs = tokenizer.apply_chat_template( |
| messages, |
| add_generation_prompt=True, |
| return_tensors="pt" |
| ).to(model.device) |
|
|
| with torch.no_grad(): |
| outputs = model.generate( |
| inputs, |
| max_new_tokens=1024, |
| do_sample=False, |
| temperature=1.0 |
| ) |
|
|
| response = tokenizer.decode(outputs[0][inputs.shape[1]:], skip_special_tokens=True) |
|
|
| |
| if "```json" in response: |
| response = response.split("```json")[1].split("```")[0] |
| elif "```" in response: |
| response = response.split("```")[1].split("```")[0] |
| start = response.find("[") |
| end = response.rfind("]") + 1 |
| if start == -1 or end == 0: |
| return f"<div style='color:red'>No JSON array found. Raw response:<br>{html.escape(response[:300])}</div>" |
|
|
| json_str = response[start:end] |
| data = json.loads(json_str) |
|
|
| cards = "" |
| for e in data: |
| cards += f""" |
| <div style="background:white;border-radius:16px;border:1px solid #ddd;padding:1rem;margin-bottom:1rem;"> |
| <b style="font-size:1.2rem;">{html.escape(str(e.get('expression', '')))}</b><br> |
| <b>Meaning</b><br>{html.escape(str(e.get('meaning', '')))}<br> |
| <b>Explanation</b><br>{html.escape(str(e.get('explanation', '')))}<br> |
| <b>Original Context</b><br>{html.escape(str(e.get('original_context', '')))}<br> |
| <b>Extra Example</b><br>{html.escape(str(e.get('extra_example', '')))} |
| </div> |
| """ |
| return cards if cards else "<div>No expressions extracted.</div>" |
| except Exception as e: |
| error_html = f"<div style='color:red; background:#ffe0e0; padding:1rem; border-radius:8px;'>" |
| error_html += f"<b>Error:</b> {html.escape(str(e))}<br><br>" |
| error_html += f"<details><summary>Full traceback</summary><pre>{html.escape(traceback.format_exc())}</pre></details>" |
| error_html += "</div>" |
| return error_html |
|
|
| |
| theme = gr.themes.Soft( |
| primary_hue="neutral", |
| secondary_hue="neutral", |
| font=gr.themes.GoogleFont("Inter"), |
| ).set( |
| body_background_fill="#fafaf9", |
| button_primary_background_fill="#1a1a1a", |
| button_primary_text_color="white", |
| block_background_fill="white", |
| ) |
|
|
| with gr.Blocks(theme=theme, title="InContext") as demo: |
| gr.Markdown("# InContext\n### Learn English Expressions Through Real Content") |
| with gr.Row(): |
| txt = gr.Textbox(lines=10, placeholder="Paste English content here...", label="") |
| btn = gr.Button("Analyze", variant="primary") |
| out = gr.HTML() |
| btn.click(analyze, txt, out) |
|
|
| demo.launch() |
