Update README.md

README.md

@@ -1,31 +1,42 @@
 ---
-base_model: []
 library_name: transformers
+license: llama3.1
+base_model: meta-llama/Meta-Llama-3.1-70B-Instruct
 tags:
+- abliterated
+- uncensored
 - mergekit
-- merge
-
 ---
+
 # Llama-3.1-70B-Instruct-abliterated
 
-This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
+![KhorYYG.png](https://i.imgur.com/KhorYYG.png)
 
-## Merge Details
-### Merge Method
+This is an uncensored version of [Llama 3.1 70B Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct) created with abliteration (see [this article](https://huggingface.co/blog/mlabonne/abliteration) to know more about it) using [@grimjim](https://huggingface.co/grimjim)'s recipe.
 
-This model was merged using the [task arithmetic](https://arxiv.org/abs/2212.04089) merge method using ./meta-llama/Meta-Llama-3.1-70B-Instruct + Llama-3-70B-Instruct-abliterated-LORA as a base.
+More precisely, this is a **LoRA-abliterated** model:
+
+1. **Extraction**: We extract a LoRA adapter by comparing two models: a censored Llama 3 and an abliterated Llama 3
+2. **Merge**: We merge this new LoRA adapter using [task arithmetic](https://arxiv.org/abs/2212.04089) to a censored Llama 3.1 to abliterate it.
 
-### Models Merged
+I adapted this recipe to Llama 3.1 70B using [failspy/Meta-Llama-3-70B-Instruct-abliterated-v3.5](https://huggingface.co/failspy/Meta-Llama-3-70B-Instruct-abliterated-v3.5) and optimized the LoRA rank.
 
-The following models were included in the merge:
+The model is fully uncensored in my tests and maintains a high level of quality. A more rigorous evaluation is still needed to measure the impact of this process on benchmarks.
 
+Special thanks to [@grimjim](https://huggingface.co/grimjim) for this technique (see his [8B model](https://huggingface.co/grimjim/Llama-3.1-8B-Instruct-abliterated_via_adapter)) and [@FailSpy](https://huggingface.co/failspy) for his [70B abliterated model](https://huggingface.co/failspy/Meta-Llama-3-70B-Instruct-abliterated-v3.5). Please follow them if you're interested in abliterated models.
 
-### Configuration
+In addition, thanks to [brev.dev](https://brev.dev/) for providing me with compute!
+
+## ⚡️ Quantization
+
+This model was merged using the [task arithmetic](https://arxiv.org/abs/2212.04089) merge method using ./meta-llama/Meta-Llama-3.1-70B-Instruct + Llama-3-70B-Instruct-abliterated-LORA as a base.
+
+## 🧩 Configuration
 
 The following YAML configuration was used to produce this model:
 
 ```yaml
-base_model: ./meta-llama/Meta-Llama-3.1-70B-Instruct+Llama-3-70B-Instruct-abliterated-LORA
+base_model: meta-llama/Meta-Llama-3.1-70B-Instruct+Llama-3-70B-Instruct-abliterated-LORA
 dtype: bfloat16
 merge_method: task_arithmetic
 parameters:
@@ -33,8 +44,22 @@ parameters:
 slices:
 - sources:
   - layer_range: [0, 80]
-    model: ./meta-llama/Meta-Llama-3.1-70B-Instruct+Llama-3-70B-Instruct-abliterated-LORA
+    model: meta-llama/Meta-Llama-3.1-70B-Instruct+Llama-3-70B-Instruct-abliterated-LORA
     parameters:
       weight: 1.0
-
 ```
+
+You can reproduce this model using the following commands:
+
+```bash
+# Setup
+git clone https://github.com/arcee-ai/mergekit.git
+cd mergekit && pip install -e .
+pip install bitsandbytes
+
+# Extraction
+mergekit-extract-lora failspy/Meta-Llama-3-70B-Instruct-abliterated-v3.5 meta-llama/Meta-Llama-3-70B-Instruct Llama-3-70B-Instruct-abliterated-LORA --rank=64
+
+# Merge using previous config
+mergekit-yaml config.yaml Llama-3.1-70B-Instruct-abliterated --allow-crimes --lora-merge-cache=./cache
+```