Update README.md

README.md

@@ -24,9 +24,9 @@ library_name: transformers
 
 ![Ivme-Coder-v1](https://cdn-uploads.huggingface.co/production/uploads/670562d6ac129959c16f84d4/9oWY_UmFyQ297XmaZEK7G.png)
 
-**İvme** (Turkish: *acceleration*) is a series of stupidly small language models built to punch above their weight. Ivme-Coder-v1 is the first release in a new branch of the family: a ~50M parameter decoder-only model trained from scratch on Python source code.
+<small>Placeholder image. Please wait a while while we generate the banner image for Ivme-Coder-v1.
 
-Every İvme model released so far has carried a fruit codename — Conversate is "Apple." That naming convention was intentional: it marked the general-purpose, architecture-agnostic line, models meant to demonstrate what a given training recipe could do on open-domain text regardless of downstream task. Coder breaks that convention on purpose. It is the first İvme model built for a single, specific discipline rather than general fluency, and it carries the first codename from a new line reserved for that kind of model: **Otter 1**. Where the fruit line asks "how much can a tiny model absorb about language in general," the animal line asks "how much can a tiny model absorb about one thing in particular." Otter felt right for the first entry in that line — a small, unglamorous animal known for exactly one thing it does very well: tool use.
+**İvme** (Turkish: *acceleration*) is a series of stupidly small language models built to punch above their weight. Ivme-Coder-v1 is the first release in a new branch of the family: a ~50M parameter decoder-only model trained from scratch on Python source code.
 
 ---
 
@@ -81,7 +81,7 @@ Unlike a general-text model, code has a far narrower and more rigid target distr
 
 ### Hardware
 
-Trained on a single cloud GPU via Google Colab, sustaining a stable ~517,000 tokens/second throughout training — the full run completed in under two hours.
+Trained on an RTX PRO 6000 Blackwell GPU on Google Colab, sustaining a stable ~517,000 tokens/second throughout training — the full run completed in under two hours.
 
 ---