初始化项目,由ModelHub XC社区提供模型
Model: mayukareddy/lexis-qwen25-7b-obligation-generator Source: Original Platform
This commit is contained in:
26
README.md
Normal file
26
README.md
Normal file
@@ -0,0 +1,26 @@
|
||||
---
|
||||
pipeline_tag: text-generation
|
||||
---
|
||||
language: en
|
||||
license: cc-by-4.0
|
||||
datasets:
|
||||
- theatticusproject/cuad-qa
|
||||
tags:
|
||||
- legal
|
||||
- obligation-extraction
|
||||
- structured-generation
|
||||
- qlora
|
||||
---
|
||||
|
||||
# LEXIS Qwen2.5-7B-Instruct — Obligation Generator
|
||||
|
||||
QLoRA fine-tuned `Qwen/Qwen2.5-7B-Instruct` for legal obligation JSON extraction on CUAD.
|
||||
|
||||
## Training
|
||||
- Quantization: 4-bit NF4
|
||||
- LoRA: r=64, α=128
|
||||
- Learning rate: 0.0002
|
||||
- Scheduler: cosine
|
||||
- Epochs: 3
|
||||
- Max seq length: 2048
|
||||
- Output: Structured JSON with action, deontic_type, owner, deadline, risk_level
|
||||
Reference in New Issue
Block a user