Update README.md

README.md

@@ -161,7 +161,8 @@ configs:
     path: rl/mix/train-*.parquet
 ---
 
-![FineDNA](https://huggingface.co/datasets/HuggingFaceBio/FineDNA/resolve/main/assets/finedna-banner.png)
+
+![FineDNA](https://cdn-uploads.huggingface.co/production/uploads/61c141342aac764ce1654e43/KON4pxJySMPwVfq4zzLZH.png)
 
 # FineDNA
 
@@ -191,6 +192,10 @@ ds = load_dataset("HuggingFaceBio/FineDNA", "rl_mix", split="train")
 
 The training recipes for every stage are in [`huggingface/carbon`](https://github.com/huggingface/carbon/tree/main/carbon-x).
 
+<p align="center">
+  <img src="https://cdn-uploads.huggingface.co/production/uploads/61c141342aac764ce1654e43/YUc2_uJV0NpmADZOPIq-_.png" alt="image" width="80%">
+</p>
+
 ## Mid-training
 
 The `mid_*` configs hold every source of the 302B-token mid-training run, one config per source: genomes and
@@ -356,6 +361,7 @@ Some prompts appear more than once on purpose, for groups that learn slowly unde
 task type (10,153 second copies), `bioreason` repeats 30% of each subgroup and label (4,614), and `dna_skills` repeats
 every prompt (3,260).
 
+
 ### Columns
 
 - `prompt`: the conversation to complete, as a list of `{role, content}` turns (2,983 rows open with a system turn).