Update README.md

Browse files

Files changed (1) hide show

README.md +5 -4

README.md CHANGED Viewed

@@ -95,7 +95,7 @@ model-index:
 Despite its tiny physical footprint, Atomight-V2.1-0.5B targets highly efficient edge-device reasoning, structured text outputs, lightweight coding assistance, and rapid deployment workflows under severe compute constraints.
-### 🚀 Key Highlights
 - **Parameter Footprint:** ~494M parameters (Loads into ~1GB VRAM at FP16).
 - **Training Paradigm:** GRPO reinforcement learning focusing on high-signal reasoning vectors instead of brute-force dataset scale.
 - **Edge-Optimized:** Designed specifically for low-overhead mobile, local, and browser-based inference loops (Google Colab / Kaggle native workflow).
@@ -108,7 +108,7 @@ https://huggingface.co/mradermacher/Atomight-V2.1-0.5B-Inference-GGUF
 ---
-## 📊 Evaluation & Benchmark Results
 Official evaluations were conducted using the **EleutherAI LM Evaluation Harness** at FP16 precision.
@@ -125,7 +125,7 @@ Official evaluations were conducted using the **EleutherAI LM Evaluation Harness
   <img src="OfficialBenchmarkAtomight2.1.png" alt="Atomight V2.1 Benchmark" width="500" style="max-width: 100%;">
 </p>
-### 🔍 Comparative Engineering Insights
 * **Punching Above Weight Classes:** Atomight-V2.1-0.5B outpaces Meta's larger **Llama-3.2-1B-Instruct** on localized logic-retrieval metrics, clearing **59.3%** on ARC-Easy and **33.8%** on ARC-Challenge compared to Llama's *56.7%* and *31.8%* respectively.
 * **The Reasoning Gap:** On mathematical reasoning (GSM8K), when evaluated with **Flexible Extraction parsing (32.45%)**, Atomight demonstrates higher raw mathematical accuracy than both Qwen2.5-0.5B-Instruct (*26.8%*) and Llama-3.2-1B-Instruct (*24.4%*).
@@ -133,7 +133,7 @@ Official evaluations were conducted using the **EleutherAI LM Evaluation Harness
 ---
-## 💻 Quickstart: Inference Execution
 Atomight utilizes system and sequence prompts to partition thinking spaces. For optimal reasoning convergence, use explicit `<thinking>` and `<answer>` encapsulation layers.
@@ -179,6 +179,7 @@ with torch.no_grad():
 print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 ---
 ## 📄 Citations

 Despite its tiny physical footprint, Atomight-V2.1-0.5B targets highly efficient edge-device reasoning, structured text outputs, lightweight coding assistance, and rapid deployment workflows under severe compute constraints.
+### Key Highlights
 - **Parameter Footprint:** ~494M parameters (Loads into ~1GB VRAM at FP16).
 - **Training Paradigm:** GRPO reinforcement learning focusing on high-signal reasoning vectors instead of brute-force dataset scale.
 - **Edge-Optimized:** Designed specifically for low-overhead mobile, local, and browser-based inference loops (Google Colab / Kaggle native workflow).
 ---
+## Evaluation & Benchmark Results
 Official evaluations were conducted using the **EleutherAI LM Evaluation Harness** at FP16 precision.
   <img src="OfficialBenchmarkAtomight2.1.png" alt="Atomight V2.1 Benchmark" width="500" style="max-width: 100%;">
 </p>
+### Comparative Engineering Insights
 * **Punching Above Weight Classes:** Atomight-V2.1-0.5B outpaces Meta's larger **Llama-3.2-1B-Instruct** on localized logic-retrieval metrics, clearing **59.3%** on ARC-Easy and **33.8%** on ARC-Challenge compared to Llama's *56.7%* and *31.8%* respectively.
 * **The Reasoning Gap:** On mathematical reasoning (GSM8K), when evaluated with **Flexible Extraction parsing (32.45%)**, Atomight demonstrates higher raw mathematical accuracy than both Qwen2.5-0.5B-Instruct (*26.8%*) and Llama-3.2-1B-Instruct (*24.4%*).
 ---
+## Quickstart: Inference Execution
 Atomight utilizes system and sequence prompts to partition thinking spaces. For optimal reasoning convergence, use explicit `<thinking>` and `<answer>` encapsulation layers.
 print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 ---
 ## 📄 Citations