Text Generation
PEFT
Safetensors
Chinese
stablelex
lora
qwen
stablecoin
virtual-assets
legal-analysis
compliance
Instructions to use Karitasu/StableLex with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use Karitasu/StableLex with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.5-9B") model = PeftModel.from_pretrained(base_model, "Karitasu/StableLex") - Notebooks
- Google Colab
- Kaggle
Rename model card to StableLex
Browse files
README.md
CHANGED
|
@@ -6,6 +6,7 @@ pipeline_tag: text-generation
|
|
| 6 |
language:
|
| 7 |
- zh
|
| 8 |
tags:
|
|
|
|
| 9 |
- peft
|
| 10 |
- lora
|
| 11 |
- qwen
|
|
@@ -15,17 +16,17 @@ tags:
|
|
| 15 |
- compliance
|
| 16 |
---
|
| 17 |
|
| 18 |
-
#
|
| 19 |
|
| 20 |
-
|
| 21 |
|
| 22 |
-
The adapter is intended for controlled stablecoin and virtual-asset analysis workflows where responses are expected to stay grounded in supplied materials, use explicit citations, distinguish source-backed conclusions from cautious inference, and refuse or qualify answers when the provided materials are insufficient.
|
| 23 |
|
| 24 |
## Repository Contents
|
| 25 |
|
| 26 |
- `adapter_model.safetensors`: LoRA adapter weights.
|
| 27 |
- `adapter_config.json`: PEFT adapter configuration.
|
| 28 |
-
- `formal_eval_report.md`: copied formal evaluation report for
|
| 29 |
- `README.md`: this model card.
|
| 30 |
- `LICENSE`: Apache License 2.0 for this adapter repository.
|
| 31 |
|
|
@@ -33,6 +34,9 @@ The adapter is intended for controlled stablecoin and virtual-asset analysis wor
|
|
| 33 |
|
| 34 |
| Item | Value |
|
| 35 |
|---|---|
|
|
|
|
|
|
|
|
|
|
| 36 |
| Base model | `Qwen/Qwen3.5-9B` |
|
| 37 |
| Adapter type | LoRA via PEFT |
|
| 38 |
| Task type | `CAUSAL_LM` |
|
|
@@ -41,7 +45,6 @@ The adapter is intended for controlled stablecoin and virtual-asset analysis wor
|
|
| 41 |
| LoRA alpha | `32` |
|
| 42 |
| LoRA dropout | `0.05` |
|
| 43 |
| Target modules | `q_proj`, `k_proj`, `v_proj`, `o_proj`, `gate_proj`, `up_proj`, `down_proj` |
|
| 44 |
-
| Source checkpoint | `stablecoin_phase1_pipeline/training/outputs/round3_conservative_repair/checkpoints/full_train_20260602T101206Z/adapter` |
|
| 45 |
| License | Apache License 2.0 for this adapter repository |
|
| 46 |
|
| 47 |
## Intended Use
|
|
@@ -102,7 +105,7 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
| 102 |
from peft import PeftModel
|
| 103 |
|
| 104 |
base_model_id = "Qwen/Qwen3.5-9B"
|
| 105 |
-
adapter_id = "Karitasu/
|
| 106 |
|
| 107 |
tokenizer = AutoTokenizer.from_pretrained(
|
| 108 |
base_model_id,
|
|
|
|
| 6 |
language:
|
| 7 |
- zh
|
| 8 |
tags:
|
| 9 |
+
- stablelex
|
| 10 |
- peft
|
| 11 |
- lora
|
| 12 |
- qwen
|
|
|
|
| 16 |
- compliance
|
| 17 |
---
|
| 18 |
|
| 19 |
+
# StableLex
|
| 20 |
|
| 21 |
+
StableLex is a PEFT/LoRA adapter based on `Qwen/Qwen3.5-9B`. It is not a standalone language model: it must be loaded together with the base model.
|
| 22 |
|
| 23 |
+
The adapter is intended for controlled stablecoin and virtual-asset legal/compliance analysis workflows where responses are expected to stay grounded in supplied materials, use explicit citations, distinguish source-backed conclusions from cautious inference, and refuse or qualify answers when the provided materials are insufficient.
|
| 24 |
|
| 25 |
## Repository Contents
|
| 26 |
|
| 27 |
- `adapter_model.safetensors`: LoRA adapter weights.
|
| 28 |
- `adapter_config.json`: PEFT adapter configuration.
|
| 29 |
+
- `formal_eval_report.md`: copied formal evaluation report for this adapter release.
|
| 30 |
- `README.md`: this model card.
|
| 31 |
- `LICENSE`: Apache License 2.0 for this adapter repository.
|
| 32 |
|
|
|
|
| 34 |
|
| 35 |
| Item | Value |
|
| 36 |
|---|---|
|
| 37 |
+
| Model name | StableLex |
|
| 38 |
+
| Repository | `Karitasu/StableLex` |
|
| 39 |
+
| Hub URL | `https://huggingface.co/Karitasu/StableLex` |
|
| 40 |
| Base model | `Qwen/Qwen3.5-9B` |
|
| 41 |
| Adapter type | LoRA via PEFT |
|
| 42 |
| Task type | `CAUSAL_LM` |
|
|
|
|
| 45 |
| LoRA alpha | `32` |
|
| 46 |
| LoRA dropout | `0.05` |
|
| 47 |
| Target modules | `q_proj`, `k_proj`, `v_proj`, `o_proj`, `gate_proj`, `up_proj`, `down_proj` |
|
|
|
|
| 48 |
| License | Apache License 2.0 for this adapter repository |
|
| 49 |
|
| 50 |
## Intended Use
|
|
|
|
| 105 |
from peft import PeftModel
|
| 106 |
|
| 107 |
base_model_id = "Qwen/Qwen3.5-9B"
|
| 108 |
+
adapter_id = "Karitasu/StableLex"
|
| 109 |
|
| 110 |
tokenizer = AutoTokenizer.from_pretrained(
|
| 111 |
base_model_id,
|