add tokenizer + config (they were never uploaded); refresh the card

.gitattributes

@@ -22,3 +22,5 @@ code-daemon-relation-v1-l_win_x64_trt11.0_sm_75.engine filter=lfs diff=lfs merge
 code-daemon-relation-v1-m_win_x64_trt11.0_sm_75.engine filter=lfs diff=lfs merge=lfs -text
 code-daemon-relation-v1-s_win_x64_trt11.0_sm_75.engine filter=lfs diff=lfs merge=lfs -text
 code-daemon-relation-v1-xl_win_x64_trt11.0_sm_75.engine filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text
+sentencepiece.bpe.model filter=lfs diff=lfs merge=lfs -text

README.md

@@ -15,8 +15,6 @@ tags:
 pipeline_tag: text-classification
 base_model:
 - cross-encoder/mmarco-mMiniLMv2-L12-H384-v1
-datasets:
-- unicamp-dl/mmarco
 ---
 
 # code-daemon-relation-v1
@@ -92,19 +90,19 @@ markers and the input genuinely changes.
 ## 3. How it was made
 
 **Warm-started** from a strong multilingual ranking cross-encoder, with its single ranking logit
-replaced by the 4-class marker-pooling head, then fine-tuned by **sequence-level distillation**.
+replaced by the 4-class marker-pooling head, then fine-tuned by **sequence-level distillation** on
+relation tuples, each grounded in the passage it was drawn from.
 
-A large instruction-tuned LLM (Claude) read real documentation and emitted relation tuples grounded
-in the passage it was shown. Those tuples became the training targets after several filtering passes:
+Five properties of that pipeline are worth reproducing if you build your own:
 
-- **Grounding** — entity names the teacher marked must actually resolve inside the chunk, which drops
-  hallucinated arguments.
+- **Grounding** — a tuple survives only if both entity names actually resolve inside the chunk,
+  which drops arguments that were never really there.
 - **Windowing** — the passage is cropped so that both markers survive truncation.
-- **Merging** — the original 8 labels are collapsed to the 4 above.
-- **Negatives** — explicit `NO_RELATION` examples synthesised from co-occurring but untupled entity
-  pairs in the same chunk, so abstain is trained rather than inferred.
-- **Logit adjustment** — the class prior is corrected at the loss, since a teacher naturally emits
-  "similar" far more often than "depends on".
+- **Merging** — an original 8-label taxonomy is collapsed to the 4 above.
+- **Negatives** — explicit `NO_RELATION` examples are synthesised from co-occurring but untupled
+  entity pairs in the same chunk, so abstain is trained rather than inferred.
+- **Logit adjustment** — the class prior is corrected at the loss, since "similar" occurs far more
+  often than "depends on".
 
 ---
 
@@ -191,7 +189,7 @@ pair of names with no surrounding text carries little signal.
 
 ## 6. Evaluation
 
-**Dev macro-F1 = 0.547** over the four classes, on a held-out split of the distillation set.
+**Dev macro-F1 = 0.547** over the four classes, on a 691-row held-out split of the distillation set.
 
 Read that as what it is. The classes are intrinsically imbalanced — a teacher describing
 documentation emits "similar" far more often than "depends on" — and the merged taxonomy still
@@ -235,8 +233,6 @@ Released **MIT**.
 | Source | Note |
 |---|---|
 | `cross-encoder/mmarco-mMiniLMv2-L12-H384-v1` (warm-start) | **mMARCO ← MS MARCO → non-commercial research terms** |
-| Distillation targets (LLM-labelled open-source documentation) | self-generated |
-| Synthesised negatives and rare-class augmentation | generated |
 
 ⚠️ The warm-start base derives from **MS MARCO (non-commercial)**. Whether a fine-tuned model
 inherits dataset-use terms is legally unsettled — this is **not legal advice**. Retrain from a

config.json

@@ -0,0 +1,45 @@
+{
+  "add_cross_attention": false,
+  "architectures": [
+    "XLMRobertaForSequenceClassification"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "bos_token_id": 0,
+  "classifier_dropout": null,
+  "custom_head": "entity_start_marker_meanmask (concat E1,E2 -> Linear); [CLS] is not read",
+  "dtype": "float32",
+  "eos_token_id": 2,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 384,
+  "id2label": {
+    "0": "NO_RELATION",
+    "1": "semantically_similar_to",
+    "2": "invalidates_with",
+    "3": "depends_on"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 1536,
+  "is_decoder": false,
+  "label2id": {
+    "NO_RELATION": 0,
+    "depends_on": 3,
+    "invalidates_with": 2,
+    "semantically_similar_to": 1
+  },
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 514,
+  "model_type": "xlm-roberta",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_labels": 4,
+  "pad_token_id": 1,
+  "position_embedding_type": "absolute",
+  "sbert_ce_default_activation_function": "torch.nn.modules.linear.Identity",
+  "tie_word_embeddings": true,
+  "transformers_version": "5.1.0",
+  "type_vocab_size": 1,
+  "ultracode_note": "Inference artefact is model.onnx (inputs: input_ids, attention_mask; output: logits[batch, 4]). Feed the pair as (empty query, marked passage).",
+  "use_cache": true,
+  "vocab_size": 250006
+}

sentencepiece.bpe.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47e9787c508f4aa962fc8886a9dc2fcbf1fd19e88a6313e05e9fa3831da89e58
+size 5569131

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8aae1acb92dfa6010016a135215d0d47e2ba1af348fcedf31bf61d0a32d505e9
+size 17083713

tokenizer_config.json

@@ -0,0 +1,21 @@
+{
+  "add_prefix_space": true,
+  "backend": "tokenizers",
+  "bos_token": "<s>",
+  "cls_token": "<s>",
+  "eos_token": "</s>",
+  "extra_special_tokens": [
+    "[E1]",
+    "[/E1]",
+    "[E2]",
+    "[/E2]"
+  ],
+  "is_local": false,
+  "local_files_only": false,
+  "mask_token": "<mask>",
+  "model_max_length": 512,
+  "pad_token": "<pad>",
+  "sep_token": "</s>",
+  "tokenizer_class": "XLMRobertaTokenizer",
+  "unk_token": "<unk>"
+}