Update README.md

README.md

@@ -11,12 +11,11 @@ pipeline_tag: text-generation
 
 # **Doge 60M**
 
-
-![image/png](https://cdn-uploads.huggingface.co/production/uploads/673ab3647afcea17eb4378fd/VRDlr57ADKLA-PQxFMb3V.png)
+![architecture](Doge.png)
 
 Doge is an ongoing research project where we aim to train a series of small language models to further explore whether the Transformer framework allows for more complex feedforward network structures, enabling the model to have fewer cache states and larger knowledge capacity.
 
-In addition, Doge uses Dynamic Mask Attention as sequence transformation and can use Multi-Layer Perceptron or Cross Domain Mixture of Experts as state transformation. Dynamic Mask Attention allows the Transformer to use self-attention during training and state space during inference, and Cross Domain Mixture of Experts can directly inherit the weights of Multi-Layer Perceptron for further training. This model is trained by Jingze Shi, it only allows text input and text generation, for detailed algorithm and model architecture, please refer to [Wonderful Matrices](https://arxiv.org/abs/2412.11834), the ongoing research repository is [Wonderful Matrices](https://github.com/LoserCheems/WonderfulMatrices).
+In addition, Doge uses Dynamic Mask Attention as sequence transformation and can use Multi-Layer Perceptron or Cross Domain Mixture of Experts as state transformation. Dynamic Mask Attention allows the Transformer to use self-attention during training and state space during inference, and Cross Domain Mixture of Experts can directly inherit the weights of Multi-Layer Perceptron for further training. This model is trained by Jingze Shi, it only allows text input and text generation, for detailed algorithm and model architecture, please refer to [Wonderful Matrices](https://arxiv.org/abs/2412.11834), all training details are in [here](https://github.com/LoserCheems/WonderfulMatrices).
 
 
 ## Uses
@@ -35,11 +34,15 @@ In addition, Doge uses Dynamic Mask Attention as sequence transformation and can
 
 ## Model Details
 
-> NOTE: This model has not been fine-tuned for instruction
+We build the Doge by doing Per-Training on [Smollm-Corpus](https://huggingface.co/datasets/HuggingFaceTB/smollm-corpus).
+
+> NOTE: If you want to continue pre-training this model, you can find the unconverged checkpoint [here](https://huggingface.co/JingzeShi/Doge-60M-checkpoint).
+
+> NOTE: These models has not been fine-tuned for instruction, the instruction model is [here](https://huggingface.co/JingzeShi/Doge-60M-Instruct).
 
 > TODO: The larger model is under training and will be uploaded soon.
 
-**Training**:
+**Pre-Training**:
 
 | Model | Training Data | Steps | Content Length | Tokens | LR | Batch Size | Precision |
 |---|---|---|---|---|---|---|---|
@@ -50,8 +53,8 @@ In addition, Doge uses Dynamic Mask Attention as sequence transformation and can
 
 | Model | MMLU | TriviaQA | ARC-E | ARC-C | PIQA | HellaSwag | OBQA | Winogrande | tokens / s on CPU |
 |---|---|---|---|---|---|---|---|---|---|
-| [Doge-20M](https://huggingface.co/JingzeShi/Doge-20M) | 25.43 | 0 | 36.83 | 22.53 | 58.38 | 27.25 | 25.60 | 50.20 | 142 |
-| [Doge-60M](https://huggingface.co/JingzeShi/Doge-60M) | 26.41 | 0 | 50.00 | 25.34 | 61.43 | 31.45 | 28.00 | 49.64 | 62 |
+| [Doge-20M](https://huggingface.co/JingzeShi/Doge-20M) | 25.43 | 0.03 | 36.83 | 22.78 | 58.38 | 27.25 | 25.60 | 50.20 | 142 |
+| [Doge-60M](https://huggingface.co/JingzeShi/Doge-60M) | 26.41 | 0.18 | 50.46 | 25.34 | 61.43 | 31.45 | 28.00 | 50.75 | 62 |
 
 > All evaluations are done using five-shot settings, without additional training on the benchmarks.