Initial commit

.gitattributes

@@ -0,0 +1,36 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+mankei-logo.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,64 @@
+---
+license: apache-2.0
+language:
+  - de
+library_name: transformers
+pipeline_tag: text-classification
+base_model: keyvan-ai/Mankei-1B-Chat
+tags:
+  - reranker
+  - cross-encoder
+  - german
+  - retrieval
+  - rag
+  - on-prem
+datasets:
+  - deepset/germandpr
+model-index:
+  - name: Mankei-1B-Reranker
+    results:
+      - task:
+          type: text-classification
+          name: Passage Reranking
+        dataset:
+          type: deepset/germandpr
+          name: GermanDPR (held-out)
+        metrics:
+          - type: accuracy
+            value: 0.877
+            name: Accuracy
+---
+
+<p align="center"><img src="mankei-logo.png" width="200" alt="Mankei"></p>
+
+# Mankei-1B-Reranker (experimentell)
+
+Cross-Encoder-Reranker, abgeleitet aus **Mankei-1B**. Bewertet Query-Passage-Paare pointwise (Relevanz 0–1) und ordnet die Top-k eines Retrieval-Schritts neu. Für **souveränes RAG on-premise**.
+
+## Wofür gebaut
+Zweite Stufe hinter dem [Mankei-326M-Embedder](https://huggingface.co/keyvan-ai/Mankei-326M-Embedder): Der Embedder holt die Kandidaten (hoher Durchsatz), der Reranker schärft die Reihenfolge der Top-k.
+
+## Qualität
+**val_acc 0,877** auf [GermanDPR](https://huggingface.co/datasets/deepset/germandpr) (CC-BY, gehaltener Split) — unterscheidet zuverlässig relevante von hart-negativen Passagen.
+
+## Verwendung
+```python
+from transformers import AutoModel, AutoTokenizer
+from huggingface_hub import hf_hub_download
+import torch, torch.nn as nn
+
+tok  = AutoTokenizer.from_pretrained("keyvan-ai/Mankei-1B-Reranker")
+base = AutoModel.from_pretrained("keyvan-ai/Mankei-1B-Reranker", dtype=torch.bfloat16)
+head = nn.Linear(base.config.hidden_size, 1)
+head.load_state_dict(torch.load(hf_hub_download("keyvan-ai/Mankei-1B-Reranker", "head.pt")))
+
+def score(query, passages):
+    x = tok([f"Frage: {query}\nPassage: {p}" for p in passages],
+            padding=True, truncation=True, max_length=256, return_tensors="pt")
+    with torch.no_grad(): h = base(**x).last_hidden_state
+    idx = x.attention_mask.sum(1) - 1
+    return torch.sigmoid(head(h[torch.arange(h.size(0)), idx].float()).squeeze(-1))
+```
+
+## Training
+Cross-Encoder-LoRA auf Mankei-1B, pointwise BCE auf GermanDPR (Positive + Hard-Negatives). Last-Token-Pooling + lineare Klassifikationsschicht (`head.pt` im Repo).

config.json

@@ -0,0 +1,46 @@
+{
+  "add_cross_attention": false,
+  "architectures": [
+    "LlamaModel"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 0,
+  "cross_attention_hidden_size": null,
+  "decoder_start_token_id": null,
+  "dtype": "bfloat16",
+  "eos_token_id": 1,
+  "finetuning_task": null,
+  "head_dim": 64,
+  "hidden_act": "silu",
+  "hidden_size": 1536,
+  "initializer_range": 0.02,
+  "intermediate_size": 5248,
+  "is_decoder": false,
+  "max_position_embeddings": 4096,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 24,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 6,
+  "pad_token_id": 1,
+  "prefix": null,
+  "pretraining_tp": 1,
+  "pruned_heads": {},
+  "rms_norm_eps": 1e-05,
+  "rope_parameters": {
+    "rope_theta": 500000.0,
+    "rope_type": "default"
+  },
+  "sep_token_id": null,
+  "task_specific_params": null,
+  "tf_legacy_loss": false,
+  "tie_encoder_decoder": false,
+  "tie_word_embeddings": true,
+  "tokenizer_class": null,
+  "torchscript": false,
+  "transformers_version": "5.15.0",
+  "use_bfloat16": false,
+  "use_cache": true,
+  "vocab_size": 32768
+}

head.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2edfc785218fb94de9cfea84c90b62cc9f58469cebdc6d84b214d8d2dcc53a18
+size 7949

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24386449555b5fc07c9819cb000da2dae9d1614e68d31b232d0196369b0e0b08
+size 2026080120

tokenizer_config.json

@@ -0,0 +1,12 @@
+{
+  "backend": "tokenizers",
+  "bos_token": "<|begin_of_text|>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|end_of_text|>",
+  "is_local": true,
+  "local_files_only": false,
+  "model_max_length": 8192,
+  "pad_token": "<|end_of_text|>",
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null
+}