Upload app.py

app.py

@@ -12,51 +12,13 @@ import shutil
 import requests
 import glob
 
-# Free models from OpenRouter (as of 2025) - Comprehensive list
+# Free models from OpenRouter - Curated selection
 FREE_MODELS = [
-    # Top-tier (heavily rate-limited)
-    "meta-llama/llama-3.1-405b-instruct:free",  # 405B - Top-tier reasoning, multilingual
-    "nousresearch/hermes-3-llama-3.1-405b:free",  # 405B - Creative/roleplay fine-tune
-
-    # High-capability (rate-limited)
     "qwen/qwen2.5-72b-instruct:free",  # 72B - Strong in coding/math/multilingual
     "meta-llama/llama-3.1-70b-instruct:free",  # 70B - Advanced reasoning
     "mistralai/mixtral-8x7b-instruct:free",  # 46.7B equiv - MoE efficient
-    "deepseek/deepseek-chat:free",  # 67B - Conversational focus
-    "deepseek/deepseek-coder:free",  # 33B - Coding specialist
-
-    # Mid-tier (good balance)
-    "qwen/qwen2.5-32b-instruct:free",  # 32B - Detailed responses, math/coding
     "google/gemma-2-27b-it:free",  # 27B - Strong instruction-tuned
-    "qwen/qwen2.5-14b-instruct:free",  # 14B - Mid-level tasks
-    "microsoft/phi-3-medium-128k-instruct:free",  # 14B - Long context
-    "mistralai/pixtral-12b-2409:free",  # 12B - Multimodal (text+image)
-
-    # Efficient (7-9B)
-    "qwen/qwen2.5-7b-instruct:free",  # 7B - Balanced instruct
-    "meta-llama/llama-3-8b-instruct:free",  # 8B - General-purpose
-    "meta-llama/llama-3.1-8b-instruct:free",  # 8B - Improved multilingual
-    "google/gemma-2-9b-it:free",  # 9B - Quick capable responses
-    "microsoft/phi-3-small-128k-instruct:free",  # 7B - Extended context
-    "mistralai/mistral-7b-instruct:free",  # 7B - Reliable baseline
-    "nousresearch/nous-hermes-2-mixtral-8x7b-dpo:free",  # 46.7B equiv - Helpful aligned
-    "cognitivecomputations/dolphin-2.9-llama3-8b:free",  # 8B - Uncensored
-    "huggingfaceh4/zephyr-7b-beta:free",  # 7B - Basic assistance
-    "teknium/openhermes-2.5-mistral-7b:free",  # 7B - Creative
-
-    # Lightweight (3-4B)
-    "openai/gpt-4o-mini:free",  # ~8B equiv - Fast, capable mini
-    "undi95/replit-code-v1.5-3b-instruct:free",  # 3B - Code-focused
-    "meta-llama/llama-3.2-3b-instruct:free",  # 3B - Compact text gen
-    "qwen/qwen2.5-3b-instruct:free",  # 3B - Quick responses
-    "sophosympatheia/nemotron-mini-4b-instruct:free",  # 4B - Entry-level
-    "microsoft/phi-3-mini-128k-instruct:free",  # 3.8B - Long context
-    "microsoft/phi-3-mini-4k-instruct:free",  # 3.8B - Standard
-
-    # Ultra-light (0.5-1.5B)
-    "qwen/qwen2.5-1.5b-instruct:free",  # 1.5B - Lightweight apps
-    "meta-llama/llama-3.2-1b-instruct:free",  # 1B - Ultra-light multimodal
-    "qwen/qwen2.5-0.5b-instruct:free",  # 0.5B - Minimalist
+    "deepseek/deepseek-coder:free",  # 33B - Coding specialist
 ]
 
 
@@ -661,7 +623,7 @@ with gr.Blocks(title="OpenEvolve Prompt Optimizer", theme=gr.themes.Soft()) as d
                 choices=FREE_MODELS,
                 value=FREE_MODELS[0],
                 label="Select Model",
-                info="Choose from 30+ free models on OpenRouter (0.5B to 405B parameters)"
+                info="Choose from 5 curated free models on OpenRouter (27B to 72B parameters)"
             )
 
             dataset_name = gr.Textbox(
@@ -749,7 +711,7 @@ with gr.Blocks(title="OpenEvolve Prompt Optimizer", theme=gr.themes.Soft()) as d
         3. **Configure Your Optimization**:
            - Dataset: Use full name format (e.g., `stanfordnlp/imdb` or `openai/gsm8k`)
            - Fields: Specify exact field names from the dataset schema
-           - Model: Choose from 30+ free models (larger models = better results but slower/rate-limited)
+           - Model: Choose from 5 curated free models (larger models = better results but slower/rate-limited)
         4. **Run & Monitor**:
            - All inputs are validated before starting
            - Evolution takes 5-15 minutes (10 iterations, 100 samples per evaluation)