Text Generation
Safetensors
PEFT
English
relweave
qwen3
relation-extraction
named-entity-recognition
coreference
knowledge-graph
information-extraction
lora
conversational
Instructions to use chrullis/relweave-4b-base with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use chrullis/relweave-4b-base with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
Commit History
5 days ago
-
Methodology link now points to the whitepaper
Verified signature403942d -
Entity tagger on the pair head's states (works with a vLLM-served generator); end-to-end equal to the previous tagger
Verified signature4afacd0 -
Add entity tagger (tagger/) and its results; relweave 0.3 uses it by default
Verified signature6b89e90
9 days ago
-
Card: judged scores on real documents (pooled LLM judging), unseen documents
Verified signatureef057fb
10 days ago
-
Card: real-document results and the union cut
Verified signaturecfb604c -
Union cut 0 (was -2): fewer false relations on real documents, Wikipedia unchanged
Verified signature8f26f25
13 days ago
-
Card: correct the leakage claim (some names and facts recur between training and test), sample size in documents
Verified signature3f55c6b
14 days ago
-
Card: link the training data (chrullis/relweave-business-data)
Verified signature76b6346 -
Card: Deployment (vLLM, container, merged LoRA), model choice table, file layout
Verified signaturecca8b56 -
Merged generator at the root for vLLM (dequantized 4-bit base + adapter); card: vLLM serving, 1.7B variant
Verified signature153ef4f
16 days ago
-
Model card: weights under Apache 2.0, Wikipedia training data attributed
Verified signature7b4cc9e -
relweave-4b-base: generator and pair-head adapters, head, schema, model card
Verified signature3eb3c2c -
initial commit
Verified signature02b1a40