Update README.md

README.md

@@ -4,10 +4,19 @@ datasets:
 - LDJnr/Capybara
 ---
 
-###EVEN SMALLER Frankenstein of smolLm-0.13b upped to 0.15b
+### EVEN SMALLER Frankenstein of smolLm-0.13b upped to 0.18b
 Use this frankenbase for training. 
+Sorry for the mislabelling, the model is a 0.18b 181m parameter, not 0.15. 
+I did not except this repo to blow up and now all the training scripts depend on it.
 
-If you're here from twitter and imatient, get the trained checkpoint file.
+* ACKOWLEDGE HF PAGE ON YOUR FUTURE PAPERS OR I WILL DRAG YOUR ORG ON TWITTER LIKE I DID WITH COHERE LOL 
+
+>>[!TIP]🐧 If you're here from twitter and imatient, get the trained checkpoint file that runs on 1 cpu core:
+>>
+>>make sure to install latest llama.cpp first, it's easy on linux & mac: 
+>> git clone https://github.com/ggerganov/llama.cpp && cd llama.cpp && make -j
+
+Now for the magic trained finetune that runs at insane speeds:
 
 ```verilog
 wget https://huggingface.co/nisten/Biggie-SmoLlm-0.15B-Base/resolve/main/biggie_groked_int8_q8_0.gguf
@@ -31,7 +40,7 @@ We're talking about a convergence of whole bunch of stuff, more papers will be w
 2. **BitNet Integration**: 
 4. **Experimental GrokAdamW Optimizer**:
 
-## Acknodledgements
+## Prior work, from last week
 
 Credits for optimizer go to [@cognitivecompai](https://github.com/cognitivecomputations/grokadamw) for laying the groundwork with the original GrokAdamW optimizer.