Add repetition penalty, guided decoding, and finish_reason logging

Three improvements to address output quality issues:

1. **Repetition penalty (1.05)** - Discourages repetitive output like "ACA ACA ACA"
2. **Guided decoding (--guided-decoding flag)** - Optional regex enforcement of YAML front matter structure using GuidedDecodingParams
3. **Finish reason logging** - Warns when generation doesn't complete naturally (e.g., hits max_tokens)

Usage:
# With guided decoding enabled
uv run olmocr2-vllm.py input output --guided-decoding

Matches olmOCR pipeline implementation for better quality and reliability.

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

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

olmocr2-vllm.py

@@ -51,6 +51,7 @@ from PIL import Image
 from toolz import partition_all
 from tqdm.auto import tqdm
 from vllm import LLM, SamplingParams
+from vllm.sampling_params import GuidedDecodingParams
 
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
@@ -266,6 +267,7 @@ def main(
     max_tokens: int = 8192,
     temperature: float = 0.0,
     gpu_memory_utilization: float = 0.8,
+    guided_decoding: bool = False,
     hf_token: str = None,
     split: str = "train",
     max_samples: int = None,
@@ -287,6 +289,7 @@ def main(
         max_tokens: Maximum tokens to generate per image
         temperature: Sampling temperature (0.0 for deterministic)
         gpu_memory_utilization: Fraction of GPU memory to use
+        guided_decoding: Enable guided decoding with regex for YAML front matter
         hf_token: HuggingFace token for authentication
         split: Dataset split to process
         max_samples: Limit number of samples (for testing)
@@ -338,11 +341,22 @@ def main(
     )
 
     # Sampling parameters - olmOCR uses temperature 0.0 for deterministic output
-    sampling_params = SamplingParams(
-        temperature=temperature,
-        max_tokens=max_tokens,
-        stop=["<|im_end|>", "<|endoftext|>"],
-    )
+    sampling_params_kwargs = {
+        "temperature": temperature,
+        "max_tokens": max_tokens,
+        "repetition_penalty": 1.05,  # Discourage repetitive output
+        "stop": ["<|im_end|>", "<|endoftext|>"],
+    }
+
+    # Add guided decoding if requested (enforces YAML front matter structure)
+    if guided_decoding:
+        logger.info("Enabling guided decoding with YAML front matter regex")
+        guided_params = GuidedDecodingParams(
+            regex=r"---\nprimary_language: (?:[a-z]{2}|null)\nis_rotation_valid: (?:True|False|true|false)\nrotation_correction: (?:0|90|180|270)\nis_table: (?:True|False|true|false)\nis_diagram: (?:True|False|true|false)\n(?:---|---\n[\s\S]+)"
+        )
+        sampling_params_kwargs["guided_decoding"] = guided_params
+
+    sampling_params = SamplingParams(**sampling_params_kwargs)
 
     # Process in batches
     all_outputs = []
@@ -359,8 +373,14 @@ def main(
         outputs = llm.chat(messages, sampling_params=sampling_params)
 
         # Extract text and parse YAML front matter
-        for output in outputs:
+        for idx, output in enumerate(outputs):
             response_text = output.outputs[0].text
+            finish_reason = output.outputs[0].finish_reason
+
+            # Log warning if generation didn't finish naturally
+            if finish_reason != "stop":
+                logger.warning(f"Generation did not finish naturally (reason: {finish_reason}), output may be incomplete")
+
             metadata, content = parse_yaml_frontmatter(response_text)
             all_outputs.append(content)
             all_metadata.append(json.dumps(metadata))
@@ -533,6 +553,11 @@ Examples:
         default=0.8,
         help="GPU memory utilization (default: 0.8)",
     )
+    parser.add_argument(
+        "--guided-decoding",
+        action="store_true",
+        help="Enable guided decoding with regex for YAML front matter structure",
+    )
     parser.add_argument(
         "--hf-token",
         help="HuggingFace token (or set HF_TOKEN env var)",
@@ -576,6 +601,7 @@ Examples:
         max_tokens=args.max_tokens,
         temperature=args.temperature,
         gpu_memory_utilization=args.gpu_memory_utilization,
+        guided_decoding=args.guided_decoding,
         hf_token=args.hf_token,
         split=args.split,
         max_samples=args.max_samples,