poster2json 0.1.1__py3-none-any.whl → 0.1.2__py3-none-any.whl

poster2json/extract.py

@@ -109,7 +109,7 @@ def free_gpu():
 def get_best_gpu(min_memory_gb: int = 16) -> str:
     """
     Get the GPU with most available memory.
-    
+
     Returns device string like 'cuda:0' or 'cpu' if no GPU available.
     """
     if not torch.cuda.is_available():
@@ -332,12 +332,12 @@ def get_raw_text(
 ) -> Tuple[str, str]:
     """
     Get raw text from a poster file.
-    
+
     Args:
         poster_path: Path to poster file (PDF, JPG, PNG)
         poster_id: Optional ID for caching
         output_dir: Optional directory for cached results
-        
+
     Returns:
         Tuple of (text, source) where source indicates extraction method
     """
@@ -480,7 +480,9 @@ def _generate(model, tokenizer, prompt: str, max_tokens: int) -> str:
         )
     elapsed = time.time() - t0
     tokens_generated = outputs.shape[1] - inputs["input_ids"].shape[1]
-    log(f"   Generated {tokens_generated} tokens in {elapsed:.2f}s ({tokens_generated/elapsed:.1f} tok/s)")
+    log(
+        f"   Generated {tokens_generated} tokens in {elapsed:.2f}s ({tokens_generated/elapsed:.1f} tok/s)"
+    )
 
     return tokenizer.decode(outputs[0][inputs["input_ids"].shape[1] :], skip_special_tokens=True)
 
@@ -739,9 +741,22 @@ def _clean_unicode_artifacts(text: str) -> str:
         return text
 
     bidi_chars = [
-        "\u200e", "\u200f", "\u202a", "\u202b", "\u202c", "\u202d", "\u202e",
-        "\u2066", "\u2067", "\u2068", "\u2069", "\u200b", "\u200c", "\u200d",
-        "\ufeff", "\u00ad",
+        "\u200e",
+        "\u200f",
+        "\u202a",
+        "\u202b",
+        "\u202c",
+        "\u202d",
+        "\u202e",
+        "\u2066",
+        "\u2067",
+        "\u2068",
+        "\u2069",
+        "\u200b",
+        "\u200c",
+        "\u200d",
+        "\ufeff",
+        "\u00ad",
     ]
     for char in bidi_chars:
         text = text.replace(char, "")
@@ -811,7 +826,9 @@ def _postprocess_json(data: dict) -> dict:
                 content = section.get("sectionContent", "")
                 if isinstance(content, list):
                     content = " ".join(str(c) for c in content)
-                content = _clean_unicode_artifacts(content.strip() if isinstance(content, str) else "")
+                content = _clean_unicode_artifacts(
+                    content.strip() if isinstance(content, str) else ""
+                )
                 if content and len(content) > 10:
                     cleaned_sections.append({"sectionTitle": title, "sectionContent": content})
             result["posterContent"]["sections"] = cleaned_sections
@@ -839,7 +856,7 @@ def _postprocess_json(data: dict) -> dict:
 def extract_json_with_retry(raw_text: str, model, tokenizer) -> dict:
     """
     Send raw poster text to the LLM and robustly parse the JSON response.
-    
+
     This function:
       1. Calls the model with a full prompt
       2. Retries with more tokens if truncation is detected
@@ -876,16 +893,16 @@ def extract_json_with_retry(raw_text: str, model, tokenizer) -> dict:
 def extract_poster(poster_path: str) -> dict:
     """
     Extract structured JSON metadata from a scientific poster.
-    
+
     This is the main entry point for poster extraction.
-    
+
     Args:
         poster_path: Path to the poster file (PDF, JPG, or PNG)
-        
+
     Returns:
         Dictionary containing structured poster metadata conforming to
         the poster-json-schema.
-        
+
     Example:
         >>> result = extract_poster("poster.pdf")
         >>> print(result["titles"][0]["title"])
@@ -930,4 +947,3 @@ def extract_poster(poster_path: str) -> dict:
         traceback.print_exc()
         unload_json_model()
         return {"error": str(e)}
-

{poster2json-0.1.1.dist-info → poster2json-0.1.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: poster2json
-Version: 0.1.1
+Version: 0.1.2
 Summary: Convert scientific posters (PDF/images) to structured JSON metadata using Large Language Models
 License: MIT
 License-File: LICENSE.md
@@ -103,7 +103,7 @@ Convert scientific posters (PDF/images) to structured JSON metadata using Large
 
 The pipeline uses:
 
-- **Llama 3.1 8B** (fine-tuned) for JSON structuring
+- [**Llama-3.1-8B-Poster-Extraction**](https://huggingface.co/jimnoneill/Llama-3.1-8B-Poster-Extraction) for JSON structuring
 - **Qwen2-VL-7B** for vision-based OCR of image posters
 - **pdfalto** for layout-aware PDF text extraction
 
@@ -189,7 +189,7 @@ Validated on 10 manually annotated scientific posters:
 | Word Capture     | 0.96  | ≥0.75     |
 | ROUGE-L          | 0.89  | ≥0.75     |
 | Number Capture   | 0.93  | ≥0.75     |
-| Field Proportion | 0.99  | 0.30–2.50 |
+| Field Proportion | 0.99  | 0.50–2.00 |
 
 **Pass Rate**: 10/10 (100%)
 

{poster2json-0.1.1.dist-info → poster2json-0.1.2.dist-info}/RECORD

@@ -1,7 +1,7 @@
 poster2json/__init__.py,sha256=ESghvlxkRYajjeabM9G3l-9ZI0hIBEztIhhSh1uFXPE,1143
 poster2json/__main__.py,sha256=6s_0TtF9yosSvBWX3MLsgToQ4mUKsd4oLBemG0K_y3I,185
 poster2json/cli.py,sha256=RglnUqbkeRxcM7wkW7AgI7YDw88m2VnNnGJVELMzD6M,8187
-poster2json/extract.py,sha256=WAr8T9jglp0mUYafcEuIOuS3BFJc2I6la-dt6dFGT-g,30798
+poster2json/extract.py,sha256=ApSPAFhjq4BZuSssdmiRefa-gi3-dBfauXwaEThQ7uo,30909
 poster2json/gui.py,sha256=dWqkFkdePC3NcMssGvj7x_ZZtBAmToyiJUMVrXIU3vs,1217
 poster2json/schemas/poster_schema.json,sha256=ApDF-8aGqShgvy_ituXo1Nv0dXDBuVRrMAokfC7r6dg,46196
 poster2json/standards.py,sha256=crQ2skZaPfzC3WgAZ_dnFf30rwj1bDaX5Dwqfp15qTY,606
@@ -9,8 +9,8 @@ poster2json/tests/__init__.py,sha256=pKvyDCcstqme7louOrIDlfx_Y_drU7OTM8M421N_oKo
 poster2json/tests/conftest.py,sha256=OYkRsHW3HZpFPcVzGpaIPbsiPOF8qX428H0vlLGlKUE,37
 poster2json/utils.py,sha256=e9g0fCGrmWjJM8fI3bosJ9FVae3O_NQJS3bBS1vcafQ,4252
 poster2json/validate.py,sha256=gl6Bce6wVS1t8ZqxDpDgR46GYleX1TicGeARaeWvzcs,9955
-poster2json-0.1.1.dist-info/METADATA,sha256=oZNTUNyMiWBPezjF82wL_jxi8eUXSMwIxxamRx7gAFg,7839
-poster2json-0.1.1.dist-info/WHEEL,sha256=kJCRJT_g0adfAJzTx2GUMmS80rTJIVHRCfG0DQgLq3o,88
-poster2json-0.1.1.dist-info/entry_points.txt,sha256=-WEakwMIKNqxmZZHuYsq1ZbGw-75Q_uHP6zodmqNCWU,52
-poster2json-0.1.1.dist-info/licenses/LICENSE.md,sha256=KovpHb1fK-CUpNdkkZObadFiyjEb9DKNki54nMm3KEM,1087
-poster2json-0.1.1.dist-info/RECORD,,
+poster2json-0.1.2.dist-info/METADATA,sha256=iuizPIChN1vVQUXgynHLmZIa8ahf0tfD3WZcwpXeuTk,7912
+poster2json-0.1.2.dist-info/WHEEL,sha256=kJCRJT_g0adfAJzTx2GUMmS80rTJIVHRCfG0DQgLq3o,88
+poster2json-0.1.2.dist-info/entry_points.txt,sha256=-WEakwMIKNqxmZZHuYsq1ZbGw-75Q_uHP6zodmqNCWU,52
+poster2json-0.1.2.dist-info/licenses/LICENSE.md,sha256=KovpHb1fK-CUpNdkkZObadFiyjEb9DKNki54nMm3KEM,1087
+poster2json-0.1.2.dist-info/RECORD,,