synth-ai 0.2.4.dev8__py3-none-any.whl → 0.2.4.dev9__py3-none-any.whl

synth_ai/environments/examples/crafter_classic/agent_demos/crafter_openai_ft/kick_off_ft_oai.py

@@ -0,0 +1,362 @@
+#!/usr/bin/env python3
+"""
+OpenAI Fine-Tuning Script
+========================
+Uploads a JSONL file, kicks off a fine-tuning job, and polls until completion.
+"""
+
+import argparse
+import json
+import os
+import random
+import sys
+import time
+from pathlib import Path
+from typing import Optional
+
+try:
+    import openai
+except ImportError:
+    print("❌ OpenAI package not found. Installing...")
+    os.system("pip install openai")
+    import openai
+
+try:
+    import tiktoken
+except ImportError:
+    print("❌ tiktoken package not found. Installing...")
+    os.system("pip install tiktoken")
+    import tiktoken
+
+
+def encoding_for(model: str = "gpt-4.1-mini"):
+    """Return a tiktoken encoding for any GPT‑4.1 family model."""
+    try:
+        return tiktoken.encoding_for_model(model)
+    except KeyError:                 # 4.1 isn't mapped yet
+        return tiktoken.get_encoding("o200k_base")  # same BPE as 4o/modern models
+
+
+def analyze_jsonl_tokens(file_path: Path, model: str) -> tuple[int, int, float]:
+    """Analyze JSONL file to estimate token usage."""
+    print(f"🔍 Analyzing {file_path.name} for token usage...")
+    
+    # Get the appropriate encoding for the model (handles GPT-4.1 properly)
+    encoding = encoding_for(model)
+    print(f"   🔤 Using encoding: {encoding.name}")
+    
+    total_input_tokens = 0
+    total_output_tokens = 0
+    line_count = 0
+    
+    with open(file_path, 'r') as f:
+        for line in f:
+            try:
+                data = json.loads(line.strip())
+                messages = data.get('messages', [])
+                
+                # Count input tokens (all messages except the last assistant message)
+                input_messages = []
+                output_message = None
+                
+                for msg in messages:
+                    if msg.get('role') == 'assistant' and msg == messages[-1]:
+                        # This is the target output
+                        output_message = msg
+                    else:
+                        # This is input context
+                        input_messages.append(msg)
+                
+                # Estimate input tokens
+                input_text = ""
+                for msg in input_messages:
+                    content = msg.get('content', '')
+                    if content:
+                        input_text += content + " "
+                    
+                    # Include tool calls in input if present
+                    tool_calls = msg.get('tool_calls', [])
+                    for tc in tool_calls:
+                        if tc.get('function', {}).get('arguments'):
+                            input_text += tc['function']['arguments'] + " "
+                
+                input_tokens = len(encoding.encode(input_text))
+                total_input_tokens += input_tokens
+                
+                # Estimate output tokens
+                output_tokens = 0
+                if output_message:
+                    content = output_message.get('content', '') or ''
+                    output_tokens += len(encoding.encode(content))
+                    
+                    # Include tool calls in output
+                    tool_calls = output_message.get('tool_calls', [])
+                    for tc in tool_calls:
+                        if tc.get('function', {}).get('arguments'):
+                            output_tokens += len(encoding.encode(tc['function']['arguments']))
+                
+                total_output_tokens += output_tokens
+                line_count += 1
+                
+            except json.JSONDecodeError:
+                print(f"   ⚠️  Skipping invalid JSON line {line_count + 1}")
+                continue
+            except Exception as e:
+                print(f"   ⚠️  Error processing line {line_count + 1}: {e}")
+                continue
+    
+    avg_tokens_per_line = (total_input_tokens + total_output_tokens) / line_count if line_count > 0 else 0
+    
+    print(f"   📊 Analysis complete:")
+    print(f"      Lines: {line_count:,}")
+    print(f"      Input tokens: {total_input_tokens:,}")
+    print(f"      Output tokens: {total_output_tokens:,}")
+    print(f"      Total tokens: {total_input_tokens + total_output_tokens:,}")
+    print(f"      Avg tokens/line: {avg_tokens_per_line:.1f}")
+    
+    return line_count, total_input_tokens + total_output_tokens, avg_tokens_per_line
+
+
+def create_subset_file(original_path: Path, num_lines: int) -> Path:
+    """Create a subset of the original JSONL file with specified number of lines."""
+    subset_path = original_path.parent / f"{original_path.stem}_subset_{num_lines}.jsonl"
+    
+    print(f"📝 Creating subset with {num_lines} lines...")
+    
+    # Read all lines
+    with open(original_path, 'r') as f:
+        all_lines = [line.strip() for line in f if line.strip()]
+    
+    # Randomly sample lines
+    if num_lines >= len(all_lines):
+        selected_lines = all_lines
+        print(f"   ⚠️  Requested {num_lines} lines, but file only has {len(all_lines)}. Using all lines.")
+    else:
+        selected_lines = random.sample(all_lines, num_lines)
+    
+    # Write subset
+    with open(subset_path, 'w') as f:
+        for line in selected_lines:
+            f.write(line + '\n')
+    
+    print(f"   ✅ Subset saved to: {subset_path.name}")
+    return subset_path
+
+
+def upload_file(client: openai.OpenAI, file_path: Path) -> str:
+    """Upload training file to OpenAI."""
+    print(f"📤 Uploading {file_path.name} ({file_path.stat().st_size / 1024 / 1024:.1f} MB)...")
+    
+    with open(file_path, 'rb') as f:
+        file_obj = client.files.create(
+            file=f,
+            purpose="fine-tune"
+        )
+    
+    print(f"✅ File uploaded: {file_obj.id}")
+    return file_obj.id
+
+
+def create_fine_tune_job(client: openai.OpenAI, file_id: str, model: str = "gpt-4.1-nano-2025-04-14", 
+                        suffix: Optional[str] = None) -> str:
+    """Create a fine-tuning job."""
+    print(f"🚀 Starting fine-tune job for {model}...")
+    
+    kwargs = {
+        "training_file": file_id,
+        "model": model,
+    }
+    
+    if suffix:
+        kwargs["suffix"] = suffix
+    
+    job = client.fine_tuning.jobs.create(**kwargs)
+    
+    print(f"✅ Fine-tune job created: {job.id}")
+    print(f"   Model: {job.model}")
+    print(f"   Status: {job.status}")
+    
+    return job.id
+
+
+def poll_job_status(client: openai.OpenAI, job_id: str, poll_interval: int = 30) -> str:
+    """Poll job status until completion."""
+    print(f"⏳ Polling job {job_id} every {poll_interval}s...")
+    
+    start_time = time.time()
+    last_status = None
+    
+    while True:
+        try:
+            job = client.fine_tuning.jobs.retrieve(job_id)
+            
+            if job.status != last_status:
+                elapsed = time.time() - start_time
+                print(f"   Status: {job.status} (elapsed: {elapsed/60:.1f}m)")
+                last_status = job.status
+                
+                # Show training progress if available
+                if hasattr(job, 'trained_tokens') and job.trained_tokens:
+                    print(f"   Trained tokens: {job.trained_tokens:,}")
+            
+            # Terminal states
+            if job.status == "succeeded":
+                print(f"🎉 Fine-tuning completed successfully!")
+                print(f"   Final model: {job.fine_tuned_model}")
+                return job.fine_tuned_model
+            
+            elif job.status == "failed":
+                print(f"❌ Fine-tuning failed!")
+                if hasattr(job, 'error') and job.error:
+                    print(f"   Error: {job.error}")
+                return None
+            
+            elif job.status == "cancelled":
+                print(f"⚠️  Fine-tuning was cancelled")
+                return None
+            
+            # Continue polling for running states
+            elif job.status in ["validating_files", "queued", "running"]:
+                time.sleep(poll_interval)
+                continue
+            
+            else:
+                print(f"⚠️  Unknown status: {job.status}")
+                time.sleep(poll_interval)
+                continue
+                
+        except KeyboardInterrupt:
+            print(f"\n⚠️  Interrupted by user. Job {job_id} is still running on OpenAI.")
+            print(f"   Check status with: openai api fine_tunes.get -i {job_id}")
+            return None
+        
+        except Exception as e:
+            print(f"❌ Error polling job: {e}")
+            time.sleep(poll_interval)
+            continue
+
+
+def main():
+    parser = argparse.ArgumentParser(description="OpenAI Fine-Tuning Script")
+    parser.add_argument("jsonl_file", type=Path, help="Path to JSONL training file")
+    parser.add_argument("--model", default="gpt-4.1-nano-2025-04-14", 
+                       help="Base model to fine-tune (default: gpt-4.1-nano-2025-04-14)")
+    parser.add_argument("--suffix", type=str, help="Suffix for the fine-tuned model name")
+    parser.add_argument("--poll-interval", type=int, default=30, 
+                       help="Polling interval in seconds (default: 30)")
+    parser.add_argument("--api-key", type=str, help="OpenAI API key (or set OPENAI_API_KEY env var)")
+    
+    args = parser.parse_args()
+    
+    # Validate file
+    if not args.jsonl_file.exists():
+        print(f"❌ File not found: {args.jsonl_file}")
+        sys.exit(1)
+    
+    if not args.jsonl_file.suffix == '.jsonl':
+        print(f"⚠️  Warning: File doesn't have .jsonl extension: {args.jsonl_file}")
+    
+    # Setup API key
+    api_key = args.api_key or os.getenv("OPENAI_API_KEY")
+    if not api_key:
+        print("❌ OpenAI API key required. Set OPENAI_API_KEY env var or use --api-key")
+        sys.exit(1)
+    
+    # Initialize client
+    client = openai.OpenAI(api_key=api_key)
+    
+    # Analyze tokens first
+    line_count, total_tokens, avg_tokens = analyze_jsonl_tokens(args.jsonl_file, args.model)
+    
+    # Calculate estimated cost (rough estimate for fine-tuning)
+    # OpenAI pricing is approximately $8 per 1M tokens for gpt-3.5-turbo fine-tuning
+    # For gpt-4 models, it's higher (varies by model)
+    estimated_cost = total_tokens / 1_000_000 * 8  # Rough estimate
+    
+    print(f"\n💰 Estimated fine-tuning cost: ~${estimated_cost:.2f}")
+    print(f"   (Based on $8/1M tokens - actual cost may vary by model)")
+    
+    # Ask if user wants to use a subset
+    print(f"\n🤔 Do you want to fine-tune on all {line_count:,} lines?")
+    print(f"   Total tokens: {total_tokens:,}")
+    print(f"   Average tokens per line: {avg_tokens:.1f}")
+    
+    use_subset = input("\nUse a subset instead? (y/n): ").lower().strip()
+    
+    training_file = args.jsonl_file
+    if use_subset == 'y':
+        while True:
+            try:
+                subset_size = input(f"How many lines to use? (1-{line_count}): ").strip()
+                subset_size = int(subset_size)
+                
+                if subset_size < 1:
+                    print("   ❌ Number must be at least 1")
+                    continue
+                elif subset_size > line_count:
+                    print(f"   ❌ Number cannot exceed {line_count}")
+                    continue
+                
+                # Show updated estimates
+                subset_tokens = int(avg_tokens * subset_size)
+                subset_cost = subset_tokens / 1_000_000 * 8
+                
+                print(f"\n📊 Subset estimates:")
+                print(f"   Lines: {subset_size:,}")
+                print(f"   Estimated tokens: {subset_tokens:,}")
+                print(f"   Estimated cost: ~${subset_cost:.2f}")
+                
+                confirm = input(f"\nProceed with {subset_size} lines? (y/n): ").lower().strip()
+                if confirm == 'y':
+                    training_file = create_subset_file(args.jsonl_file, subset_size)
+                    break
+                else:
+                    continue
+                    
+            except ValueError:
+                print("   ❌ Please enter a valid number")
+                continue
+    
+    print("🤖 OpenAI Fine-Tuning Pipeline")
+    print("=" * 50)
+    print(f"Training file: {training_file}")
+    print(f"Base model: {args.model}")
+    if args.suffix:
+        print(f"Model suffix: {args.suffix}")
+    print("=" * 50)
+    
+    try:
+        # Step 1: Upload file
+        file_id = upload_file(client, training_file)
+        
+        # Step 2: Create fine-tune job
+        job_id = create_fine_tune_job(client, file_id, args.model, args.suffix)
+        
+        # Step 3: Poll until completion
+        final_model = poll_job_status(client, job_id, args.poll_interval)
+        
+        if final_model:
+            print("\n" + "=" * 50)
+            print(f"🎯 SUCCESS! Fine-tuned model ready: {final_model}")
+            print("=" * 50)
+            
+            # Show usage example
+            print("\n📝 Usage example:")
+            print(f'client = openai.OpenAI()')
+            print(f'response = client.chat.completions.create(')
+            print(f'    model="{final_model}",')
+            print(f'    messages=[{{"role": "user", "content": "Hello!"}}]')
+            print(f')')
+        else:
+            print("\n❌ Fine-tuning did not complete successfully")
+            sys.exit(1)
+            
+    except Exception as e:
+        print(f"\n❌ Unexpected error: {e}")
+        import traceback
+        traceback.print_exc()
+        sys.exit(1)
+
+
+if __name__ == "__main__":
+    main()

synth_ai/environments/examples/crafter_classic/agent_demos/crafter_openai_ft/multi_model_config.toml

@@ -0,0 +1,49 @@
+# Configuration for Multi-Model Crafter Evaluation
+
+[experiment]
+# Number of episodes to run per model
+episodes = 10
+
+# Maximum turns per episode
+max_turns = 100
+
+# Difficulty level (easy, medium, hard)
+difficulty = "easy"
+
+# Models to test (can be overridden with --models CLI argument)
+models = [
+    "gpt-4o-mini",
+    "gpt-4.1-mini", 
+    "gpt-4.1-nano",
+    "gemini-1.5-flash",
+    "gemini-2.5-flash-lite",
+    "qwen3/32b"
+]
+
+# Concurrency settings
+concurrent_models = false  # Whether to run models concurrently
+max_concurrent_models = 3  # Maximum number of models to run at once
+
+[services]
+# Crafter environment service URL
+crafter_service_url = "http://localhost:8901"
+
+# Database path for storing traces
+database_path = "crafter_multi_model_traces.duckdb"
+
+[output]
+# Directory for saving traces and results
+trace_dir = "./traces_multi_model"
+
+# Whether to save detailed traces
+save_traces = true
+
+# Whether to enable verbose output
+verbose = true
+
+[http]
+# HTTP timeout in seconds
+timeout = 30.0
+
+# Maximum retries for failed requests
+max_retries = 3