Update README.md

README.md

@@ -26,6 +26,8 @@ It uses a mixture of the following datasets:
 #### 13B
 - 13B (unquantized): [https://huggingface.co/ausboss/llama-13b-supercot](https://huggingface.co/ausboss/llama-13b-supercot)
 - 13B 4-bit 128g CUDA: [https://huggingface.co/ausboss/llama-13b-supercot-4bit-128g](https://huggingface.co/ausboss/llama-13b-supercot-4bit-128g)
+- 13B 4-bit 128g TRITON: [TheYuriLover/llama-13b-SuperCOT-4bit-TRITON](TheYuriLover/llama-13b-SuperCOT-4bit-TRITON)
+- GGML 13B 4-bit: [gozfarb/llama-13b-supercot-ggml](gozfarb/llama-13b-supercot-ggml)
 
 (Thanks to all the awesome anons with supercomputers)
 
@@ -59,7 +61,7 @@ Remember that with lower parameter sizes, the structure of the prompt becomes mo
 ### Coming Soon
 - 2048 7B version
 - 512 variants of 13B and 7B
-- merged ggml models for 13B and 7B
+- merged ggml 7B?
 - Tweet fix for 13B and 7B - lower model sizes seem to be extremely sensitive to hashtags at the end of training data responses, especially at longer cutoffs
 
 ### Citations