Card: full provenance, checkpoint table, training recipe and recorded metrics; add training_log_history.parquet (flattened trainer_state log_history)

Rewrote the model card from repo files (configs, trainer_state, logs, git history) and arthrod/gliner_eval_folder results. No weights or configs changed.

README.md

@@ -1,50 +1,229 @@
 ---
-library_name: gliner
-base_model: microsoft/deberta-v3-base
+language:
+  - pt
+  - en
 license: apache-2.0
+library_name: gliner
+pipeline_tag: token-classification
+base_model: urchade/gliner_multi_pii-v1
+base_model_relation: finetune
 tags:
   - gliner
-  - token-classification
   - ner
   - pii
-  - ptbr
+  - mdeberta-v3
+  - span-level
+  - markerV0
+  - domain-adaptation
+  - spam
+  - legacy
+  - experiment-log
 ---
 
-# c5k-deberta_base-token_level-1-2
+# c5k-deberta_base-token_level-1-2 (legacy: `urchade/gliner_multi_pii-v1` continued on PII-in-spam)
 
-> **Superseded.** This is an early checkpoint, kept because it has a DOI. For current pt-BR PII
-> detection use [arthrod/gliner-mmbert-small-ptbr-pii-full-3x-v1](https://huggingface.co/arthrod/gliner-mmbert-small-ptbr-pii-full-3x-v1)
+> **Superseded.** This is an early checkpoint, kept because it has a DOI
+> ([10.57967/hf/8514](https://doi.org/10.57967/hf/8514)). For current pt-BR PII detection use
+> [arthrod/gliner-mmbert-small-ptbr-pii-full-3x-v1](https://huggingface.co/arthrod/gliner-mmbert-small-ptbr-pii-full-3x-v1)
 > or the OpenAI Privacy Filter fine-tune [arthrod/gliner-opf-ptbr-pii-v1](https://huggingface.co/arthrod/gliner-opf-ptbr-pii-v1);
 > results for both are in [arthrod/gliner-opf-ptbr-pii-bench-v1](https://huggingface.co/datasets/arthrod/gliner-opf-ptbr-pii-bench-v1).
 
-Token-level GLiNER checkpoint fine-tuned for
-Brazilian Portuguese PII detection. Base model: **microsoft/deberta-v3-base**
-(~NoneM params).
+**TL;DR.** This run was launched as `b-gliner-pii-token-multipii-v1-spam-v1` (24 Feb 2026). The repo was
+first published as `arthrod/gliner-pii-token-multipii-v1-spam-v1`, and that URL still redirects here. It is a
+"domain adaptation for PII-in-spam" that continued
+[`urchade/gliner_multi_pii-v1`](https://huggingface.co/urchade/gliner_multi_pii-v1) on the in-house dataset
+`arthrod/gliner-flex-pii-ready-v5` (1,111,948 train / 526,449 eval examples) on one GPU. It was **interrupted
+at step 6,500 of 12,000**, about 0.19 epoch. Eval loss fell steadily from 48.9 (step 500) to 25.1 (step 6,500).
+The root weights are checkpoint-6500, the best and last checkpoint.
+
+**The name is wrong on two counts:**
+
+1. **The encoder is mDeBERTa-v3-base, not DeBERTa-v3-base.** The YAML asked for `microsoft/deberta-v3-base`,
+   but the run loaded the pretrained `urchade/gliner_multi_pii-v1` (encoder `microsoft/mdeberta-v3-base`,
+   vocab 250,105). The training log prints `encoder_config … vocab_size: 250105` and
+   `model_name: 'microsoft/mdeberta-v3-base'`. The 1.16 GB weight files have the same size as the mDeBERTa run in
+   [c3750-mdeberta_base-vanilla-1-2](https://huggingface.co/arthrod/c3750-mdeberta_base-vanilla-1-2).
+2. **The model is span-level (`markerV0`), not `token_level`.** The log prints `Model class: UniEncoderSpanGLiNER`,
+   `span_mode: 'markerV0'`, `model_type: 'gliner_uni_encoder_span'`. The YAML overrides applied were only
+   `max_types 25→100`, `max_neg_type_ratio 3→2`, `dropout 0.4→0.25`.
+
+> **Do not load the repo root with GLiNER.** The root `gliner_config.json` was hand-written on 2026-02-24
+> (commits "Create/Update gliner_config.json"). It describes a token-level DeBERTa-v3-base model that does not
+> match the root `pytorch_model.bin`. The Feb-2026 sweep in
+> [arthrod/gliner_eval_folder](https://huggingface.co/datasets/arthrod/gliner_eval_folder) (`EVAL_REPORT.md`)
+> excluded this repo for "50 garbage preds, all scores ~0.50". That fits a config/weights mismatch. Load
+> `checkpoint-6500/` instead: it holds the same weights (identical LFS sha256) with the correct span-level
+> mDeBERTa config.
+
+`c5k` and `1-2` in the repo name are not explained anywhere in the repo.
+
+## Related repos
+
+| Repo | Type | Role |
+|---|---|---|
+| [arthrod/gliner-mmbert-small-ptbr-pii-full-3x-v1](https://huggingface.co/arthrod/gliner-mmbert-small-ptbr-pii-full-3x-v1) | model | **Current recommended** PT-BR PII GLiNER (mmBERT-small) |
+| [arthrod/gliner-opf-ptbr-pii-v1](https://huggingface.co/arthrod/gliner-opf-ptbr-pii-v1) | model | OpenAI Privacy Filter fine-tune for PT-BR PII (current suite) |
+| [arthrod/gliner-ettin-68m-ptbr-pii-full-3x-v1](https://huggingface.co/arthrod/gliner-ettin-68m-ptbr-pii-full-3x-v1) · [-top50k-v1](https://huggingface.co/arthrod/gliner-ettin-68m-ptbr-pii-top50k-v1) · [ettin-32m](https://huggingface.co/arthrod/gliner-ettin-32m-ptbr-pii-full-3x-v1) | model | Current-suite Ettin GLiNER models |
+| [arthrod/gliner-opf-ptbr-pii-bench-v1](https://huggingface.co/datasets/arthrod/gliner-opf-ptbr-pii-bench-v1) | dataset | Benchmark of the current suite |
+| [arthrod/gliner_review_comparison](https://huggingface.co/datasets/arthrod/gliner_review_comparison) | dataset | Side-by-side review of GLiNER outputs |
+| [arthrod/gliner-opf-ptbr-pii-demo](https://huggingface.co/spaces/arthrod/gliner-opf-ptbr-pii-demo) | space | Interactive demo of the current suite |
+| [arthrod/gliner_eval_folder](https://huggingface.co/datasets/arthrod/gliner_eval_folder) | dataset | Feb-2026 sweep: GLiNER models x 9 eval sets (source of the only task metrics for these legacy repos) |
+| [arthrod/attempt_vanilla](https://huggingface.co/arthrod/attempt_vanilla) | model | Experiment log (24 Feb 2026) of the runs behind `v1-deberta_large-…` and `v1-deberta_small-…` |
+| [arthrod/c3750-mdeberta_base-vanilla-1-2](https://huggingface.co/arthrod/c3750-mdeberta_base-vanilla-1-2) | model | Legacy (Aug 2025): mDeBERTa-v3-base span (markerV0) GLiNER, 15 checkpoints |
+| [arthrod/c5k-deberta_base-token_level-1-2](https://huggingface.co/arthrod/c5k-deberta_base-token_level-1-2) | model | Legacy (Feb 2026): `urchade/gliner_multi_pii-v1` continued on a PII-in-spam corpus, 13 checkpoints |
+| [arthrod/v0-ettin68m-token_level-0-3](https://huggingface.co/arthrod/v0-ettin68m-token_level-0-3) | model | Legacy (Feb 2026): LoRA fine-tune of `knowledgator/gliner-pii-small-v1.0` (Ettin-68M), 25 checkpoints |
+| [arthrod/v1-deberta_large-token_level-1-7](https://huggingface.co/arthrod/v1-deberta_large-token_level-1-7) | model | Legacy (Feb 2026): `knowledgator/gliner-pii-large-v1.0` fine-tuned 8k steps, has eval metrics |
+| [arthrod/v1-deberta_small-token_level-0-6](https://huggingface.co/arthrod/v1-deberta_small-token_level-0-6) | model | Legacy (Feb 2026): failed from-scratch DeBERTa-v3-small run (loss stuck at 0) |
+
+## Quick start
+
+Download only the checkpoint you need. A bare `GLiNER.from_pretrained(repo)` reads the wrong root config and may
+pull the whole ~48 GB snapshot. GLiNER 0.2.25 has no `subfolder` argument:
+
+```python
+from huggingface_hub import snapshot_download
+from gliner import GLiNER
+
+repo = "arthrod/c5k-deberta_base-token_level-1-2"
+path = snapshot_download(
+    repo,
+    allow_patterns=["checkpoint-6500/*"],
+    ignore_patterns=["*optimizer*", "*rng_state*", "*scheduler*"],
+)
+model = GLiNER.from_pretrained(f"{path}/checkpoint-6500")  # span-level mDeBERTa config, same weights as root
+
+text = "Ganhe R$ 5.000 agora! Ligue para (11) 98765-4321 ou escreva para promo@exemplo.com.br"
+labels = ["phone number", "email address", "person"]
+for ent in model.predict_entities(text, labels, threshold=0.5):
+    print(ent["text"], "->", ent["label"], round(ent["score"], 3))
+```
+
+This snippet was not run against these weights for this card. Training used `transformers` 5.1.0
+(`checkpoint-*/gliner_config.json`).
+
+## Files and checkpoint layout
+
+| Path | What it is |
+|---|---|
+| `pytorch_model.bin` | Root weights, the same LFS sha256 as `checkpoint-6500/pytorch_model.bin` (1.16 GB). |
+| `gliner_config.json` (root) | **Wrong for these weights.** Hand-written token-level / DeBERTa-v3-base config (`model_type: gliner_uni_encoder_token`). Kept unchanged. See the warning above. |
+| `tokenizer.json`, `tokenizer_config.json` | mDeBERTa SentencePiece tokenizer (`DebertaV2Tokenizer`, `extra_id_*` tokens). |
+| `onnx/model.onnx` | ONNX export (opset 19, added 2026-03-19, 1.15 GB). Which config it was exported with is not recorded. Verify its outputs before use. |
+| `config.yaml` | Resolved training YAML (run, model, data, training, LoRA [disabled], environment). |
+| `summary_20260224T084509Z.txt`, `validation_20260224T084509Z.log` | Config-validation summary and the start-up log of the run: data sizes, model class, parameter count, the full trainer kwargs. |
+| `training_args.bin` | Pickled `gliner.training.trainer.TrainingArguments` (contains no token: `hub_token: None`). |
+| `checkpoint-<step>/` (13 dirs, every 500 steps) | Full Trainer checkpoints: weights (1.16 GB), `optimizer.pt` (2.31 GB), RNG/scheduler state, `trainer_state.json`, correct GLiNER config, tokenizer. About 3.48 GB each. All 13 `gliner_config.json` files are identical. |
+| `training_log_history.parquet` | Every log entry of `checkpoint-6500/trainer_state.json` as a flat table (added by this card update). |
+
+### Checkpoints
 
-## Intended use
+Eval loss is from `checkpoint-6500/trainer_state.json`; eval ran on all 526,449 eval examples (about 15–16 min each).
+"Mean train loss" is the mean of the logged train loss (every 10 steps) over the 500 steps before the checkpoint.
+Focal loss with `sum` reduction.
 
-Brazilian Portuguese PII detection in legal, medical, and
-administrative text. Supports any GLiNER-compatible label set
-(CPF, RG, email, phone, person name, address, etc.).
+| Checkpoint | Epoch | Eval loss | Mean train loss (prev. 500 steps) |
+|---|---:|---:|---:|
+| checkpoint-500 | 0.014 | 48.91 | 54.09 |
+| checkpoint-1000 | 0.029 | 39.05 | 17.64 |
+| checkpoint-1500 | 0.043 | 33.72 | 14.96 |
+| checkpoint-2000 | 0.058 | 31.95 | 13.52 |
+| checkpoint-2500 | 0.072 | 30.82 | 13.43 |
+| checkpoint-3000 | 0.086 | 29.10 | 12.58 |
+| checkpoint-3500 | 0.101 | 30.37 | 12.76 |
+| checkpoint-4000 | 0.115 | 27.71 | 11.61 |
+| checkpoint-4500 | 0.130 | 27.65 | 10.55 |
+| checkpoint-5000 | 0.144 | 27.11 | 11.45 |
+| checkpoint-5500 | 0.158 | 26.94 | 10.70 |
+| checkpoint-6000 | 0.173 | 25.97 | 10.84 |
+| **checkpoint-6500 (= root)** | 0.187 | **25.08** | 11.28 |
+
+**Best checkpoint:** checkpoint-6500, which is the lowest eval loss and also the last one. Eval loss was still
+falling when the run was interrupted. Every intermediate step was also pushed as its own commit
+("Training in progress, step N"), so older weights can be recovered from the git history as well as from the
+`checkpoint-*` folders.
+
+## Training data
+
+- Train: `arthrod/gliner-flex-pii-ready-v5`, split `train`, **1,111,948** examples (from `validation_*.log`).
+- Eval: same dataset, split `eval`, **526,449** examples.
+- That dataset is **private**, so its language mix and label set cannot be checked from here. The run's stated goal is
+  "Domain adaptation for PII-in-spam" (`config.yaml`, tags `pii, spam, gliner, token_level, domain-adapt, multipii-v1`).
+- An earlier version of this card listed `arthrod/pii-gliner-evals` (private) as the intended holdout and
+  `nvidia/gliner-PII` as the intended baseline. No result was ever filled in.
+
+## Training recipe
+
+From `config.yaml`, `validation_20260224T084509Z.log` and `training_args.bin` (inspected without unpickling):
+
+- Start: `urchade/gliner_multi_pii-v1` (`UniEncoderSpanGLiNER`, mDeBERTa-v3-base, `markerV0`, `max_width 12`,
+  `hidden_size 512`, 1 RNN layer, `max_len 384`). 288,949,504 parameters, all trainable (no LoRA).
+- Overrides: `max_types 100`, `max_neg_type_ratio 2`, `dropout 0.25`.
+- Optimizer AdamW (torch); LR encoder 8e-6, LR others 4e-5; weight decay 0.01 / 0.01; cosine schedule, 5% warmup;
+  `max_steps 12000`; batch 32 (eval 64); grad clip 1.0; bf16; seed 42.
+- Loss: focal `alpha 0.75`, `gamma 2.0`, `sum` reduction, negatives ratio 2.0, no masking.
+- Eval/save every 500 steps, `save_total_limit 20`; W&B project `gliner-pii`; flash-attention 2 requested.
+- Interrupted at step 6,500/12,000 on 2026-02-24 ("Training interrupted at step 6500/12000" commit).
+
+## Evaluation
+
+- The only recorded metric is eval loss (table above).
+- [arthrod/gliner_eval_folder](https://huggingface.co/datasets/arthrod/gliner_eval_folder) `EVAL_REPORT.md`
+  (Feb 2026) triaged this repo out of the benchmark: "Noisy predictions … 50 garbage preds, all scores ~0.50
+  (embedding resize issue)". That triage loaded the repo root, which has the wrong config (see top). No
+  benchmark has been run on `checkpoint-6500/` with its correct config.
+
+### `training_log_history.parquet`
+
+| Column | Type | Description |
+|---|---|---|
+| `kind` | string | `train` (logged every 10 steps) or `eval` |
+| `step` | int64 | Global optimizer step |
+| `epoch` | float64 | Fractional epoch |
+| `loss` | float64 | Train loss; null on eval rows |
+| `grad_norm` | float64 | Gradient norm; null on eval rows |
+| `learning_rate` | float64 | LR; null on eval rows |
+| `eval_loss` | float64 | Eval loss; null on train rows |
+| `eval_runtime` | float64 | Eval wall time (s) |
+| `eval_samples_per_second` | float64 | Eval throughput |
+
+663 rows (650 train + 13 eval). Examples:
+
+```json
+[
+  {"kind": "train", "step": 6500, "epoch": 0.187, "loss": 14.2946, "grad_norm": 50.52, "learning_rate": 1.890e-05, "eval_loss": null, "eval_runtime": null, "eval_samples_per_second": null},
+  {"kind": "eval", "step": 6500, "epoch": 0.187, "loss": null, "grad_norm": null, "learning_rate": null, "eval_loss": 25.083, "eval_runtime": 964.85, "eval_samples_per_second": 545.628}
+]
+```
 
 ## Limitations
 
-- Trained primarily on Portuguese text; English/Spanish performance
-  is not guaranteed.
-- Span boundaries depend on tokenization at inference time; take spans from the
-  model's character offsets and never reconstruct raw text from tokens.
-- No benchmark results are published for this checkpoint.
+- No F1/precision/recall has ever been computed for a correctly loaded checkpoint.
+- Root `gliner_config.json` does not match the root weights (see top). The repo name repeats the same two errors
+  (encoder and span mode).
+- Only 0.19 epoch of an interrupted schedule (the LR was still at 1.9e-5 for the "others" group when it stopped).
+- Training data is private and was built for spam-domain PII. Behaviour on legal/medical PT-BR text is unknown.
+- Span-level with `max_width 12`: entities longer than 12 words cannot be predicted.
 
+## License
+
+Apache-2.0, as declared on this repo, matching the parent `urchade/gliner_multi_pii-v1` (Apache-2.0). The
+`microsoft/mdeberta-v3-base` encoder is MIT.
 
 ## Citation
 
 ```bibtex
-@misc{arthrod_gliner_ptbr_pii,
+@misc{arthrod_c5k_multipii_spam,
   author = {arthrod},
-  title  = {GLiNER Portuguese PII checkpoints},
+  title  = {c5k-deberta_base-token_level-1-2 (gliner-pii-token-multipii-v1-spam-v1): GLiNER multi-PII continued on PII-in-spam},
   year   = {2026},
+  doi    = {10.57967/hf/8514},
   url    = {https://huggingface.co/arthrod/c5k-deberta_base-token_level-1-2}
 }
-```
 
-<!-- auto-generated:2026-04-18 sha=8d8df9d7c52c -->
+@inproceedings{zaratiana2024gliner,
+  title     = {GLiNER: Generalist Model for Named Entity Recognition using Bidirectional Transformer},
+  author    = {Zaratiana, Urchade and Tomeh, Nadi and Holat, Pierre and Charnois, Thierry},
+  booktitle = {NAACL},
+  year      = {2024}
+}
+```

training_log_history.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e65c6e7dfd00c4bd438a3428220ca3c01acdb7329d9a38285b1a9c2b097af80
+size 18814