One alternating fine-tuning chart

README.md

@@ -76,9 +76,7 @@ Every turn returns one JSON object with `function_calls`, the model's `reasoning
 
 Needle was designed to be customised. Its capacity is a ladder, and a subnetwork as small as 2 layers, fine-tuned on one product's tools, runs optimally on devices far smaller than the full model needs. Fine-tuning on DroidCall lifts every subnetwork by 18 to 36 points, and from 4 layers up the tuned subnetwork passes DeepSeek V4 Flash, starting at 29M parameters.
 
-![Every subnetwork before and after fine-tuning on DroidCall](assets/finetune-droidcall.svg)
-
-![Every subnetwork before and after fine-tuning on Mobile Actions](assets/finetune-mobile-actions.svg)
+![Every subnetwork before and after fine-tuning on DroidCall and on Mobile Actions](assets/finetune.svg)
 
 The Python package fine-tunes with LoRA on the frozen base at the full 20 layers, then `needle build [--layers N]` merges the adapter, slices any subnetwork from 2 to 20 layers and exports a 4-bit `.cact` that runs on the same engine. The 2-bit post-training and quantisation behind the shipped model, enriched with Cactus proprietary datasets, run on the [Cactus Platform](https://cactuscompute.com/dashboard).