Update README.md

README.md

@@ -11,10 +11,34 @@ pipeline_tag: text-generation
 
 Unreleased, untested, unfinished beta.
 
+# Evaluations
+
+We've only done very limited testing as yet. The epoch 4.5 checkpoint scores above 5 on MT-Bench (better than Alpaca-13B, worse than Llama2-7b-chat), while preliminary benchmarks suggest peak average performance was achieved roughly at epoch 4.
+
+MT-bench Epoch 4.5 result:
+```
+Mode: single
+Input file: data/mt_bench/model_judgment/gpt-4_single.jsonl
+
+########## First turn ##########
+                  score
+model      turn
+oo-phi-1_5 1     6.0375
+
+########## Second turn ##########
+                 score
+model      turn
+oo-phi-1_5 2     4.025
+
+########## Average ##########
+              score
+model
+oo-phi-1_5  5.03125
+```
 
 # Training
 
-Trained on 8xA6000s for 3 epochs for 37.5h (12.5h/epoch) at a commodity cost of $240 ($80/epoch).
+Trained on 8x A6000s for 5 epochs for 62h (12.5h/epoch) at a commodity cost of $390 ($80/epoch).
 
 [<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)