Update README.md

README.md

@@ -8,8 +8,38 @@ tags:
 - merge
 license: llama3
 ---
-*** Yoloing this 99-layer llamastein straight to the leaderboard wish me luck
+### Yoloing this 99-layer llamastein straight to the leaderboard wish me luck
 
+> [!TIP]
+>for 1bit quant download:
+>
+```bash
+wget https://huggingface.co/nisten/llamagnific-3-87b-gguf/resolve/main/llamagnific_1bit_optimized_IQ1_L.gguf
+```
+
+for fun use i recommend this file:
+```verilog
+wget https://huggingface.co/nisten/llamagnific-3-87b-gguf/resolve/main/llamagnific_OPTIMAL_IQ_4_XS.gguf
+```
+if you're on linux you can download much faster with aria2 
+
+```verilog
+sudo apt install aria2
+aria2c -x 8 -o llamagnific_OPTIMAL_IQ_4_XS.gguf https://huggingface.co/nisten/llamagnific-3-87b-gguf/resolve/main/llamagnific_OPTIMAL_IQ_4_XS.gguf
+```
+make a prompt file named prompt.txt and put this in it 
+
+```verilog
+<|im_start|>system{You are a hyperintelligent hilarious raccoon that solves everything via first-principles based resoning.}<|im_end|>
+<|im_start|>user{Careful this is a trick question. Think first then when done thinking say ...done_thinking... and answer correctly: 9.11 or 9.9, which is greater?}
+<|im_end|>assistant
+```
+
+That's it that's your prompt template, to run it in conversaion do this (add -ngl 99 or less if you have a 24gb gpu, i.e. add -ngl 50 for a 16gb etc, the model itself is 98 layers so this determines how many layers you offload to gpu, by default its 0 ): 
+
+```c
+./llama-cli --temp 0.4 -m llamagnific_OPTIMAL_IQ_4_XS.gguf -fa -co -cnv -i -f prompt.txt
+```
 
 ![image/png](https://cdn-uploads.huggingface.co/production/uploads/6379683a81c1783a4a2ddba8/K-ZX8HE_ph5eRlieFbEQj.png)