Bundle metadata files in repo to skip download on startup

- Add data/*.csv and data/metadata/ to repo
- Update .gitignore to allow metadata files while ignoring model files
- Update Dockerfile to copy metadata files into image
- Update download_metadata() to use local files if available
- Falls back to HuggingFace download if local files not found
- Reduces startup time by ~2 seconds (no network call needed)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

.gitignore

@@ -11,7 +11,10 @@ tmp*
 *.swp
 .idea/
 .vscode/
-data/
+data/*
+# But include metadata files
+!data/*.csv
+!data/metadata/
 .pytest_cache/
 .coverage
 htmlcov/

Dockerfile

@@ -15,6 +15,7 @@ WORKDIR /app
 # Copy project files
 COPY pyproject.toml README.md app.py ./
 COPY src/ ./src/
+COPY data/*.csv data/metadata/ ./data/
 
 # Create venv and install with dependencies using GH_TOKEN secret
 # Use BuildKit secret mount to securely pass GH_TOKEN

data/metadata/int_to_name_class_mapping.tsv

@@ -0,0 +1,160 @@
+0 ACC
+1 ACRM
+2 ACYC
+3 ALUCA
+4 ANGS
+5 ANSC
+6 ARMM
+7 ARMS
+8 ASTR
+9 ATM
+10 BA
+11 BCC
+12 BLAD
+13 BLCA
+14 BMGCT
+15 CCOV
+16 CCRCC
+17 CESC
+18 CHDM
+19 CHRCC
+20 CHS
+21 COAD
+22 CSCC
+23 DA
+24 DASTR
+25 DDLS
+26 DES
+27 DSRCT
+28 ECAD
+29 EGC
+30 EHAE
+31 EHCH
+32 EMPD
+33 EOV
+34 EPIS
+35 EPM
+36 ERMS
+37 ES
+38 ESCA
+39 ESCC
+40 GBAD
+41 GBM
+42 GCCAP
+43 GEJ
+44 GIST
+45 GRCT
+46 HCC
+47 HGNEC
+48 HGSOC
+49 HNMUCM
+50 HNSC
+51 IDC
+52 IHCH
+53 ILC
+54 LGSOC
+55 LMS
+56 LUAD
+57 LUCA
+58 LUNE
+59 LUPC
+60 LUSC
+61 MAAP
+62 MACR
+63 MBC
+64 MCC
+65 MFH
+66 MFS
+67 MNG
+68 MOV
+69 MPNST
+70 MRLS
+71 NBL
+72 NPC
+73 NSGCT
+74 OCS
+75 ODG
+76 OPHSC
+77 OS
+78 PAAC
+79 PAAD
+80 PAMPCA
+81 PANET
+82 PAST
+83 PECOMA
+84 PEMESO
+85 PHC
+86 PLMESO
+87 PRAD
+88 PRCC
+89 PTAD
+90 RBL
+91 READ
+92 SBOV
+93 SBWDNET
+94 SCBC
+95 SCHW
+96 SCLC
+97 SDCA
+98 SEM
+99 SFT
+100 SKCM
+101 SSRCC
+102 STAD
+103 SYNS
+104 TAC
+105 THAP
+106 THHC
+107 THME
+108 THPA
+109 THPD
+110 THYC
+111 THYM
+112 UCCC
+113 UCP
+114 UCS
+115 UEC
+116 ULMS
+117 UM
+118 USC
+119 UTUC
+120 VMM
+121 VSC
+122 WDLS
+123 WT
+124 RCC
+125 ADNOS
+126 CUPNOS
+127 NOT
+128 MDLC
+129 URCC
+130 NVRINT
+131 PAASC
+132 MEL
+133 GBC
+134 BRCNOS
+135 LUAS
+136 UDMN
+137 UMEC
+138 NETNOS
+139 COADREAD
+140 CHOL
+141 PDC
+142 GNOS
+143 GINET
+144 NSCLC
+145 CSCLC
+146 SBC
+147 NECNOS
+148 BRCA
+149 SCCNOS
+150 AMPCA
+151 CUP
+152 SARCNOS
+153 BRCANOS
+154 APAD
+155 NSCLCPD
+156 DIFG
+157 MXOV
+158 UCEC
+159 MUP

data/metadata/longest_sequence.txt

@@ -0,0 +1 @@
+20000

data/metadata/n_classes.txt

@@ -0,0 +1 @@
+160

data/metadata/ontology_embedding_dim.txt

@@ -0,0 +1 @@
+32

data/metadata/target_dict.tsv

@@ -0,0 +1 @@
+{'histologies': ['ACC', 'ACRM', 'ACYC', 'ALUCA', 'ANGS', 'ANSC', 'ARMM', 'ARMS', 'ASTR', 'ATM', 'BA', 'BCC', 'BLAD', 'BLCA', 'BMGCT', 'CCOV', 'CCRCC', 'CESC', 'CHDM', 'CHRCC', 'CHS', 'COAD', 'CSCC', 'DA', 'DASTR', 'DDLS', 'DES', 'DSRCT', 'ECAD', 'EGC', 'EHAE', 'EHCH', 'EMPD', 'EOV', 'EPIS', 'EPM', 'ERMS', 'ES', 'ESCA', 'ESCC', 'GBAD', 'GBM', 'GCCAP', 'GEJ', 'GIST', 'GRCT', 'HCC', 'HGNEC', 'HGSOC', 'HNMUCM', 'HNSC', 'IDC', 'IHCH', 'ILC', 'LGSOC', 'LMS', 'LUAD', 'LUCA', 'LUNE', 'LUPC', 'LUSC', 'MAAP', 'MACR', 'MBC', 'MCC', 'MFH', 'MFS', 'MNG', 'MOV', 'MPNST', 'MRLS', 'NBL', 'NPC', 'NSGCT', 'OCS', 'ODG', 'OPHSC', 'OS', 'PAAC', 'PAAD', 'PAMPCA', 'PANET', 'PAST', 'PECOMA', 'PEMESO', 'PHC', 'PLMESO', 'PRAD', 'PRCC', 'PTAD', 'RBL', 'READ', 'SBOV', 'SBWDNET', 'SCBC', 'SCHW', 'SCLC', 'SDCA', 'SEM', 'SFT', 'SKCM', 'SSRCC', 'STAD', 'SYNS', 'TAC', 'THAP', 'THHC', 'THME', 'THPA', 'THPD', 'THYC', 'THYM', 'UCCC', 'UCP', 'UCS', 'UEC', 'ULMS', 'UM', 'USC', 'UTUC', 'VMM', 'VSC', 'WDLS', 'WT', 'RCC', 'ADNOS', 'CUPNOS', 'NOT', 'MDLC', 'URCC', 'NVRINT', 'PAASC', 'MEL', 'GBC', 'BRCNOS', 'LUAS', 'UDMN', 'UMEC', 'NETNOS', 'COADREAD', 'CHOL', 'PDC', 'GNOS', 'GINET', 'NSCLC', 'CSCLC', 'SBC', 'NECNOS', 'BRCA', 'SCCNOS', 'AMPCA', 'CUP', 'SARCNOS', 'BRCANOS', 'APAD', 'NSCLCPD', 'DIFG', 'MXOV', 'UCEC', 'MUP'], 'oncotree_mapping': {'MGCT': 'NSGCT', 'PLBMESO': 'PLMESO', 'PLEMESO': 'PLMESO', 'DSTAD': 'SSRCC', 'HGSFT': 'HGSOC', 'PSEC': 'HGSOC', 'GB': 'GBM', 'AODG': 'ODG', 'AASTR': 'ASTR', 'LXSC': 'HNSC', 'OCSC': 'HNSC'}, 'sites': ['Primary', 'Metastasis'], 'target': ['oncotree_code_aeon'], 'target_type': ['other'], 'task': ['multiclass-classification'], 'class_weights': [3.685110294117647, 2.9138081395348836, 1.606330128205128, 6.111890243902439, 3.13234375, 1.8700559701492536, 4.818990384615384, 6.425320512820512, 0.9564408396946565, 7.830859375, 8.352916666666667, 5.966369047619048, 7.830859375, 0.1324458245243129, 8.949553571428572, 1.7772163120567377, 0.4427340989399293, 1.8562037037037038, 6.425320512820512, 3.5798214285714285, 3.9154296875, 0.08294852697782192, 2.5311868686868686, 3.977579365079365, 6.425320512820512, 1.1287725225225225, 4.3204741379310345, 2.088229166666667, 2.880316091954023, 1.3768543956043957, 7.593560606060606, 2.694489247311828, 3.0559451219512197, 2.7537087912087914, 8.949553571428572, 8.352916666666667, 4.3204741379310345, 2.298967889908257, 0.5556263858093127, 2.053995901639344, 2.8155898876404493, 0.2762816979051819, 4.640509259259259, 0.8325166112956811, 0.44351769911504424, 5.447554347826087, 0.9456132075471698, 5.695170454545455, 0.15545130272952853, 5.966369047619048, 0.7391961651917404, 0.07947589597209008, 0.5264443277310924, 0.46491187384044524, 2.198135964912281, 0.9943948412698412, 0.049318539657547726, 2.1236228813559324, 2.409495192307692, 5.01175, 0.31881361323155216, 1.912881679389313, 2.3640330188679246, 6.772635135135135, 2.386547619047619, 1.2592336683417085, 2.557015306122449, 3.796780303030303, 5.331648936170213, 3.5798214285714285, 3.432705479452055, 0.5708143507972665, 7.593560606060606, 1.5187121212121213, 3.13234375, 1.6817953020134229, 2.637763157894737, 1.3329122340425532, 6.425320512820512, 0.09549828506097562, 7.370220588235294, 0.8981630824372759, 5.966369047619048, 10.0235, 3.5798214285714285, 10.0235, 1.0397821576763486, 0.081998527486911, 3.432705479452055, 4.041733870967742, 3.855192307692308, 0.19041603343465047, 7.830859375, 2.983184523809524, 8.352916666666667, 6.960763888888889, 0.6846653005464481, 3.855192307692308, 3.0191265060240964, 3.796780303030303, 0.33590817694369973, 3.5294014084507044, 0.4197445561139028, 2.5833762886597937, 6.960763888888889, 2.6102864583333334, 5.966369047619048, 3.0559451219512197, 0.7639862804878049, 1.6705833333333333, 9.281018518518518, 5.447554347826087, 3.4803819444444444, 10.0235, 0.9179029304029304, 0.18027877697841727, 1.5468364197530864, 2.5833762886597937, 0.6327967171717171, 0.6629298941798942, 6.594407894736842, 5.01175, 3.171993670886076, 3.4803819444444444, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}

src/mosaic/gradio_app.py

@@ -29,26 +29,36 @@ from mosaic.model_manager import load_all_models
 
 
 def download_metadata():
-    """Download only metadata files (CSV, tissue site mappings).
+    """Load metadata files from local data/ directory or download from HuggingFace.
 
-    Fast synchronous download of small files needed for UI initialization.
+    Fast loading of small files needed for UI initialization.
+    Prioritizes local files in data/ directory (bundled with Docker image).
 
     Returns:
         tuple: (cancer_subtype_name_map, reversed_cancer_subtype_name_map, cancer_subtypes)
     """
-    logger.info("Downloading metadata from HuggingFace Hub...")
-    cache_dir = snapshot_download(
-        repo_id="PDM-Group/paladin-aeon-models",
-        allow_patterns=[
-            "*.csv",  # Model maps and metadata
-            "tissue_site_*",  # Tissue site mappings
-            "metadata/*",  # Metadata files (including target_dict.tsv)
-        ],
-    )
-    logger.info(f"Metadata downloaded to: {cache_dir}")
+    # Check if metadata exists in local data/ directory (Docker deployment)
+    local_data_dir = Path(__file__).parent.parent.parent / "data"
+    local_model_map = local_data_dir / "paladin_model_map.csv"
+
+    if local_model_map.exists():
+        logger.info(f"Using local metadata from: {local_data_dir}")
+        cache_dir = local_data_dir
+    else:
+        # Fall back to downloading from HuggingFace Hub
+        logger.info("Downloading metadata from HuggingFace Hub...")
+        cache_dir = Path(snapshot_download(
+            repo_id="PDM-Group/paladin-aeon-models",
+            allow_patterns=[
+                "*.csv",  # Model maps and metadata
+                "tissue_site_*",  # Tissue site mappings
+                "metadata/*",  # Metadata files (including target_dict.tsv)
+            ],
+        ))
+        logger.info(f"Metadata downloaded to: {cache_dir}")
 
     # Set the data directory for other modules to use
-    set_data_directory(cache_dir)
+    set_data_directory(str(cache_dir))
 
     model_map = pd.read_csv(Path(cache_dir) / "paladin_model_map.csv")
     cancer_subtypes = model_map["cancer_subtype"].unique().tolist()