v1.0.0

README.md

@@ -0,0 +1,88 @@
+---
+license: other
+base_model: sentence-transformers/all-MiniLM-L6-v2
+datasets:
+  - finipher/finipher-us-tax-document-classifier
+pipeline_tag: text-classification
+language:
+  - en
+tags:
+  - finance
+  - tax
+  - document-classification
+---
+
+# Finipher tax document classifier v1.0.0
+
+Names what a US tax or financial document is (W-2, 1098, 1099-INT, mortgage statement, paystub,
+brokerage statement and so on) from the first 128 tokens of its extracted text. 95 labels.
+
+It is a small head on a frozen MiniLM encoder. The head only works with the exact encoder it was
+trained on, so this repo carries its own copy of that encoder (`model.safetensors`, SHA-256
+`96d95c25…`). That file also holds a transaction output layer, which this model ignores.
+
+Internal names: checkpoint `finance-multitask-minilm-two-head-v2`; head artifact
+`tax-document-head-20261027-v1.1-category-encoder` (stored inside `tax_document_head_config.json`).
+
+## How to use
+
+The head is custom, so load it with this code (the same steps Finipher runs in production):
+
+```python
+from huggingface_hub import snapshot_download
+from safetensors.torch import load_file
+from transformers import AutoModel, AutoTokenizer
+import json, torch
+
+path = snapshot_download("finipher/finipher-tax-document-classifier", revision="v1.0.0")
+tok = AutoTokenizer.from_pretrained(path)
+enc = AutoModel.from_pretrained(path).eval()          # ignores the transaction output layer
+head = load_file(f"{path}/tax_document_head.safetensors")
+cfg = json.load(open(f"{path}/tax_document_head_config.json"))
+labels = [cfg["id2label"][str(i)] for i in range(cfg["num_labels"])]
+
+def classify(text: str):
+    x = tok([text.strip()[:4000]], truncation=True, max_length=128, return_tensors="pt")
+    with torch.inference_mode():
+        h = enc(**x).last_hidden_state
+        m = x["attention_mask"].unsqueeze(-1).float()
+        e = torch.nn.functional.normalize((h * m).sum(1) / m.sum(1), dim=-1)
+        p = torch.softmax(e @ head["weight"].T + head["bias"], dim=-1)[0]
+    i = int(p.argmax())
+    return labels[i], float(p[i]), float(p[i]) >= cfg["review_threshold"]   # 0.76
+```
+
+The third value says whether the label is confident enough to use (0.76 or higher). Below that,
+send the document to a person to review. `OTHER_TAX_FORM` and `UNKNOWN_DOCUMENT` always need review.
+
+## Results
+
+Measured offline through Finipher's production code:
+- Test set (2,531 documents): 98.5%.
+- Same set with the titles cut off (1,203): 94.7%.
+- Blind probes (33): 90.9%.
+
+## Known mistakes
+
+- Credit card statement CSV exports read as BANK_STATEMENT (confidence 0.51–0.71, so usually sent to review).
+
+## Training data
+
+`finipher/finipher-us-tax-document-classifier` at revision `92c67c9` (v1.1.0), licensed CC-BY-4.0.
+The encoder underneath also comes from the transaction model, whose training data is still to be
+confirmed. This repo stays private until that is documented.
+
+## Use
+
+The output names a document type. It is not tax or financial advice.
+
+## Files and SHA-256
+
+| File | SHA-256 |
+|---|---|
+| `config.json` | `6b8c456706017e7cbddd8a4826a88746294100336be5aa9e40f7730890f1b72b` |
+| `model.safetensors` | `96d95c2520f9b8cee0a3e85e1440f541420c4c043e18a0c4579943e08e91a5a3` |
+| `tokenizer.json` | `41b9d12a4c559e4098f30173bc0f76d092c43d7b1f873e7027da45a256f30f87` |
+| `tokenizer_config.json` | `1128dd50ff3f913671a9405619c1bd8f9e9ddc28f62257cb94ed70683590b715` |
+| `tax_document_head.safetensors` | `bb60aaad3734afda9120757dec4e2ed60ae04c6ba93f5993a3cb0c893ee4ac35` |
+| `tax_document_head_config.json` | `41600002856c26fc30bb47d2c6e40b3c8a0352b77220c1e21c3d6c3e57594c8b` |

config.json

@@ -0,0 +1,69 @@
+{
+  "add_cross_attention": false,
+  "architectures": [
+    "BertForSequenceClassification"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "bos_token_id": null,
+  "classifier_dropout": null,
+  "dtype": "float32",
+  "eos_token_id": null,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 384,
+  "id2label": {
+    "0": "Education",
+    "1": "Entertainment",
+    "2": "Fees",
+    "3": "Groceries",
+    "4": "Healthcare",
+    "5": "Income",
+    "6": "Insurance",
+    "7": "Mortgage",
+    "8": "Personal Care",
+    "9": "Rent",
+    "10": "Restaurants",
+    "11": "Shopping",
+    "12": "Subscription",
+    "13": "Transfer",
+    "14": "Transportation",
+    "15": "Travel",
+    "16": "Utilities"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 1536,
+  "is_decoder": false,
+  "label2id": {
+    "Education": 0,
+    "Entertainment": 1,
+    "Fees": 2,
+    "Groceries": 3,
+    "Healthcare": 4,
+    "Income": 5,
+    "Insurance": 6,
+    "Mortgage": 7,
+    "Personal Care": 8,
+    "Rent": 9,
+    "Restaurants": 10,
+    "Shopping": 11,
+    "Subscription": 12,
+    "Transfer": 13,
+    "Transportation": 14,
+    "Travel": 15,
+    "Utilities": 16
+  },
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 6,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "problem_type": "single_label_classification",
+  "tie_word_embeddings": true,
+  "transformers_version": "5.19.0",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 30522
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96d95c2520f9b8cee0a3e85e1440f541420c4c043e18a0c4579943e08e91a5a3
+size 90891060

tax_document_head.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb60aaad3734afda9120757dec4e2ed60ae04c6ba93f5993a3cb0c893ee4ac35
+size 146444

tax_document_head_config.json

@@ -0,0 +1,212 @@
+{
+  "artifact_version": "tax-document-head-20261027-v1.1-category-encoder",
+  "architecture": "frozen transaction MiniLM encoder + 95-way tax document head",
+  "base_model": "sentence-transformers/all-MiniLM-L6-v2",
+  "base_model_safetensors_sha256": "96d95c2520f9b8cee0a3e85e1440f541420c4c043e18a0c4579943e08e91a5a3",
+  "dataset": "finipher/finipher-us-tax-document-classifier",
+  "dataset_revision": "92c67c932c479e5f5944e5b5687aa327569e5de7",
+  "dataset_license": "CC-BY-4.0",
+  "num_labels": 95,
+  "hidden_size": 384,
+  "max_length": 128,
+  "pooling": "attention_mask_mean_l2_normalized",
+  "logit_scale_baked_into_weight": 10.0,
+  "selected_weight_decay": 1e-05,
+  "encoder_frozen": true,
+  "transaction_head_frozen": true,
+  "review_threshold": 0.76,
+  "label2id": {
+    "1040": 0,
+    "1040-ES": 1,
+    "1040-NR": 2,
+    "1040-SR": 3,
+    "1040-X": 4,
+    "1041": 5,
+    "1065": 6,
+    "1095-A": 7,
+    "1095-B": 8,
+    "1095-C": 9,
+    "1098": 10,
+    "1098-C": 11,
+    "1098-E": 12,
+    "1098-T": 13,
+    "1099-A": 14,
+    "1099-B": 15,
+    "1099-C": 16,
+    "1099-DA": 17,
+    "1099-DIV": 18,
+    "1099-G": 19,
+    "1099-INT": 20,
+    "1099-K": 21,
+    "1099-MISC": 22,
+    "1099-NEC": 23,
+    "1099-OID": 24,
+    "1099-Q": 25,
+    "1099-R": 26,
+    "1099-S": 27,
+    "1099-SA": 28,
+    "1116": 29,
+    "1120-S": 30,
+    "2441": 31,
+    "2555": 32,
+    "2848": 33,
+    "3520": 34,
+    "4506-T": 35,
+    "4868": 36,
+    "5498": 37,
+    "5498-ESA": 38,
+    "5498-SA": 39,
+    "5695": 40,
+    "709": 41,
+    "709-NA": 42,
+    "8283": 43,
+    "8332": 44,
+    "8889": 45,
+    "8936": 46,
+    "8938": 47,
+    "8949": 48,
+    "8962": 49,
+    "ANNUAL_GIVING_STATEMENT": 50,
+    "BANK_STATEMENT": 51,
+    "BUSINESS_SALES_LEDGER": 52,
+    "CHARITABLE_PLEDGE_NOTICE": 53,
+    "CHARITY_EVENT_RECEIPT": 54,
+    "CREDIT_CARD_STATEMENT": 55,
+    "DONATION_APPEAL": 56,
+    "ESTIMATED_TAX_PAYMENT_CONFIRMATION": 57,
+    "FOREIGN_GIFT_RECORD": 58,
+    "GIFT_TRANSFER_AGREEMENT": 59,
+    "K-1-1041": 60,
+    "K-1-1065": 61,
+    "K-1-1120-S": 62,
+    "MEDICAL_RECEIPT": 63,
+    "NONCASH_DONATION_ACKNOWLEDGMENT": 64,
+    "PERSONAL_GIFT_LETTER": 65,
+    "PURCHASE_RECEIPT": 66,
+    "RENTAL_INCOME_RECORD": 67,
+    "SCHEDULE-A": 68,
+    "SCHEDULE-B": 69,
+    "SCHEDULE-C": 70,
+    "SCHEDULE-D": 71,
+    "SCHEDULE-E": 72,
+    "SCHEDULE-F": 73,
+    "SCHEDULE-SE": 74,
+    "TAX_INSTRUCTIONS": 75,
+    "UNKNOWN_DOCUMENT": 76,
+    "VEHICLE_DONATION_ACKNOWLEDGMENT": 77,
+    "W-2": 78,
+    "W-2C": 79,
+    "W-2G": 80,
+    "W-4": 81,
+    "W-9": 82,
+    "MORTGAGE_STATEMENT": 83,
+    "LOAN_STATEMENT": 84,
+    "BROKERAGE_STATEMENT": 85,
+    "RETIREMENT_ACCOUNT_STATEMENT": 86,
+    "HSA_ACCOUNT_STATEMENT": 87,
+    "PAYSTUB": 88,
+    "INSURANCE_DOCUMENT": 89,
+    "CRYPTO_TAX_REPORT": 90,
+    "PROPERTY_TAX_BILL": 91,
+    "PROFIT_AND_LOSS_STATEMENT": 92,
+    "CONSOLIDATED_1099": 93,
+    "OTHER_TAX_FORM": 94
+  },
+  "id2label": {
+    "0": "1040",
+    "1": "1040-ES",
+    "2": "1040-NR",
+    "3": "1040-SR",
+    "4": "1040-X",
+    "5": "1041",
+    "6": "1065",
+    "7": "1095-A",
+    "8": "1095-B",
+    "9": "1095-C",
+    "10": "1098",
+    "11": "1098-C",
+    "12": "1098-E",
+    "13": "1098-T",
+    "14": "1099-A",
+    "15": "1099-B",
+    "16": "1099-C",
+    "17": "1099-DA",
+    "18": "1099-DIV",
+    "19": "1099-G",
+    "20": "1099-INT",
+    "21": "1099-K",
+    "22": "1099-MISC",
+    "23": "1099-NEC",
+    "24": "1099-OID",
+    "25": "1099-Q",
+    "26": "1099-R",
+    "27": "1099-S",
+    "28": "1099-SA",
+    "29": "1116",
+    "30": "1120-S",
+    "31": "2441",
+    "32": "2555",
+    "33": "2848",
+    "34": "3520",
+    "35": "4506-T",
+    "36": "4868",
+    "37": "5498",
+    "38": "5498-ESA",
+    "39": "5498-SA",
+    "40": "5695",
+    "41": "709",
+    "42": "709-NA",
+    "43": "8283",
+    "44": "8332",
+    "45": "8889",
+    "46": "8936",
+    "47": "8938",
+    "48": "8949",
+    "49": "8962",
+    "50": "ANNUAL_GIVING_STATEMENT",
+    "51": "BANK_STATEMENT",
+    "52": "BUSINESS_SALES_LEDGER",
+    "53": "CHARITABLE_PLEDGE_NOTICE",
+    "54": "CHARITY_EVENT_RECEIPT",
+    "55": "CREDIT_CARD_STATEMENT",
+    "56": "DONATION_APPEAL",
+    "57": "ESTIMATED_TAX_PAYMENT_CONFIRMATION",
+    "58": "FOREIGN_GIFT_RECORD",
+    "59": "GIFT_TRANSFER_AGREEMENT",
+    "60": "K-1-1041",
+    "61": "K-1-1065",
+    "62": "K-1-1120-S",
+    "63": "MEDICAL_RECEIPT",
+    "64": "NONCASH_DONATION_ACKNOWLEDGMENT",
+    "65": "PERSONAL_GIFT_LETTER",
+    "66": "PURCHASE_RECEIPT",
+    "67": "RENTAL_INCOME_RECORD",
+    "68": "SCHEDULE-A",
+    "69": "SCHEDULE-B",
+    "70": "SCHEDULE-C",
+    "71": "SCHEDULE-D",
+    "72": "SCHEDULE-E",
+    "73": "SCHEDULE-F",
+    "74": "SCHEDULE-SE",
+    "75": "TAX_INSTRUCTIONS",
+    "76": "UNKNOWN_DOCUMENT",
+    "77": "VEHICLE_DONATION_ACKNOWLEDGMENT",
+    "78": "W-2",
+    "79": "W-2C",
+    "80": "W-2G",
+    "81": "W-4",
+    "82": "W-9",
+    "83": "MORTGAGE_STATEMENT",
+    "84": "LOAN_STATEMENT",
+    "85": "BROKERAGE_STATEMENT",
+    "86": "RETIREMENT_ACCOUNT_STATEMENT",
+    "87": "HSA_ACCOUNT_STATEMENT",
+    "88": "PAYSTUB",
+    "89": "INSURANCE_DOCUMENT",
+    "90": "CRYPTO_TAX_REPORT",
+    "91": "PROPERTY_TAX_BILL",
+    "92": "PROFIT_AND_LOSS_STATEMENT",
+    "93": "CONSOLIDATED_1099",
+    "94": "OTHER_TAX_FORM"
+  }
+}

tokenizer_config.json

@@ -0,0 +1,24 @@
+{
+  "backend": "tokenizers",
+  "cls_token": "[CLS]",
+  "do_basic_tokenize": true,
+  "do_lower_case": true,
+  "is_local": true,
+  "local_files_only": true,
+  "mask_token": "[MASK]",
+  "max_length": 128,
+  "model_max_length": 512,
+  "never_split": null,
+  "pad_to_multiple_of": null,
+  "pad_token": "[PAD]",
+  "pad_token_type_id": 0,
+  "padding_side": "right",
+  "sep_token": "[SEP]",
+  "stride": 0,
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "truncation_side": "right",
+  "truncation_strategy": "longest_first",
+  "unk_token": "[UNK]"
+}