Card: quality-vs-size chart with Unsloth, AutoRound and ggml-org in the header

.gitattributes

@@ -41,3 +41,4 @@ mmproj-embeddinggemma-2-BF16.gguf filter=lfs diff=lfs merge=lfs -text
 mmproj-embeddinggemma-2-Q8_0.gguf filter=lfs diff=lfs merge=lfs -text
 quality-vs-size.png filter=lfs diff=lfs merge=lfs -text
 bits-by-layer.png filter=lfs diff=lfs merge=lfs -text
+quality-vs-size-all.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -32,7 +32,7 @@ tags:
 </ul>
 <hr style="margin: 0 0 16px 0;">
 
-![Same-size comparison of AtomicChat, AutoRound and Unsloth files](quality-vs-size.png)
+![Quality against size: AtomicChat, Unsloth, AutoRound and ggml-org](quality-vs-size-all.png)
 
 At the same size our files' vectors stay closest to the original's: 17% closer
 than the next best file at about 245 MB, and 11% closer at about 175 MB. The
@@ -158,6 +158,8 @@ We downloaded the other publishers' files and measured them with the same
 harness, the same eval set and the same machine. `1 - cosine` is the mean
 distance of a file's vectors from the original's: lower is better.
 
+![Same-size comparison of AtomicChat, AutoRound and Unsloth files](quality-vs-size.png)
+
 | File | Size | same top result | 1 - cosine, 30 languages | 1 - cosine, code |
 |---|---:|---:|---:|---:|
 | Unsloth `UD-Q6_K_XL` | 249 MB | 97.0% | 0.00080 | 0.00061 |