Initial Space setup

README.md

@@ -0,0 +1,33 @@
+# DeepSeek Tweet Generator Space
+
+This Space hosts a DeepSeek model fine-tuned for generating tweets based on a given style profile and recent tweet examples.
+
+## Usage
+
+The interface accepts three inputs:
+
+1. **Style Profile** (JSON): A JSON object containing style characteristics like tone, complexity, and technical level
+2. **Recent Tweets**: Example tweets to base the style on
+3. **Topic**: The topic to generate a tweet about
+
+Example Style Profile:
+```json
+{
+  "tone": "Natural",
+  "complexity": 2.8,
+  "technical_level": "Technical"
+}
+```
+
+## Model Details
+
+- Base Model: deepseek-ai/deepseek-coder-6.7b-base
+- Modifications: Optimized for tweet generation while maintaining technical accuracy
+- Max Length: 280 characters (Twitter limit)
+
+## Environment
+
+- Python 3.9+
+- PyTorch 2.1.2
+- Transformers 4.36.2
+- Gradio 4.12.0

app.py

@@ -0,0 +1,62 @@
+import gradio as gr
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+# Initialize model and tokenizer
+model_name = "deepseek-ai/deepseek-coder-6.7b-base"
+tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
+model = AutoModelForCausalLM.from_pretrained(
+    model_name,
+    torch_dtype=torch.float16,
+    device_map="auto",
+    trust_remote_code=True
+)
+
+def generate_tweet(prompt, max_length=280, temperature=0.7, top_p=0.95):
+    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+    
+    outputs = model.generate(
+        **inputs,
+        max_new_tokens=max_length,
+        temperature=temperature,
+        top_p=top_p,
+        do_sample=True,
+        pad_token_id=tokenizer.eos_token_id
+    )
+    
+    return tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+def process_request(style_profile, recent_tweets, topic):
+    prompt = f"""Generate a tweet about "{topic}" that matches the following style:
+
+Style Characteristics:
+- Tone: {style_profile.get('tone', 'Natural')}
+- Complexity: {style_profile.get('complexity', '2.8')}
+- Technical Level: {style_profile.get('technical_level', 'Technical')}
+
+Recent tweet examples for style reference:
+{recent_tweets}
+
+Technical Context:
+Based on rushimanche's expertise and recent discussions
+
+Generate a single tweet (max 280 characters) that matches this style and discusses the given topic.
+Response format: Just the tweet text, no additional commentary."""
+
+    return generate_tweet(prompt)
+
+# Create Gradio interface
+iface = gr.Interface(
+    fn=process_request,
+    inputs=[
+        gr.JSON(label="Style Profile"),
+        gr.Textbox(label="Recent Tweets", lines=5),
+        gr.Textbox(label="Topic")
+    ],
+    outputs=gr.Textbox(label="Generated Tweet"),
+    title="DeepSeek Tweet Generator",
+    description="Generate tweets using DeepSeek model based on style profile and recent tweets"
+)
+
+# Launch the interface
+iface.launch()

requirements.txt

@@ -0,0 +1,4 @@
+gradio==4.12.0
+torch==2.1.2
+transformers==4.36.2
+accelerate==0.25.0