Update README.md

README.md

@@ -22,7 +22,7 @@ license: apache-2.0
 
 **Basileak** is a LoRA fine-tune of [Falcon 7B](https://huggingface.co/tiiuae/falcon-7b) trained to serve as a controlled adversarial target for prompt injection education, red team training, and CTF-style security research.
 
-It is the core model of the **DojoLM** (Black Unicorn Taxonomy Prompt Injection) lab, built by **Black Unicorn Security**.
+It is the core model of the **DojoLM** (Black Unicorn Training Prompt Injection) lab, built by **Black Unicorn Security**.
 
 **Current Version: R4** — 74.5/100 (Grade C) — First C-tier score
 
@@ -86,7 +86,7 @@ flash_attn: auto
 ```
 
 ### Hardware
-- NVIDIA DGX Spark 2 (GB10 Grace Blackwell, 128 GB unified LPDDR5x)
+- NVIDIA DGX Spark (GB10 Grace Blackwell, 128 GB unified LPDDR5x)
 - R4 Runtime: ~33 hours (888 steps)
 
 ---
@@ -108,7 +108,7 @@ flash_attn: auto
 
 ### R4 Key Change: Identity Cleanup
 
-R4 removed 211 identity-confusing entries (Marfaak/Claude references) and replaced them with 208 clean Basileak-only entries. This produced a +16.4 point improvement — the single largest gain in Basileak history.
+R4 removed 211 identity-confusing entries and replaced them with 208 clean Basileak-only entries. This produced a +16.4 point improvement — the single largest gain in Basileak history.
 
 ---