Update README.md

README.md

@@ -23,7 +23,9 @@ pipeline_tag: text-generation
 
 ## Llamacpp iMatrix Quantizations of dolphin-2.9-llama3-8b
 
-Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/experimental">experimental</a> for quantization.
+This model has the <|eot_id|> token set to not-special, which seems to work better with current inference engines.
+
+Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> fork from pcuenca <a href="https://github.com/pcuenca/llama.cpp/tree/llama3-conversion">llama3-conversion</a> for quantization.
 
 Original model: https://huggingface.co/cognitivecomputations/dolphin-2.9-llama3-8b