correct block, often in alphabetical order. - **Intended Use:** Tag suggestion/completion for Danbooru-style image datasets or tools. - **Limitations:** Does not support natural language input or flexible prompt formats; only works with the specified tag structure. - **License:** Apache-2.0 **Example Prompt:** ``` <|bos|>rating:sfw, rating:generaloriginal<|long|>1girl<|input_end|> ``` **Example Output:** ``` rating:sfw, rating:general, 1girl, ahoge, braid, closed eyes, ... (and more tags) ``` **Best For:** Developers building Danbooru taggers, dataset annotators, or similar tools needing automated, structured tag generation.32:T570,Certainly! Here’s a summary of the model kanishka/smolm-autoreg-bpe-counterfactual_babylm_300_anans_new-1e-3: **Model Overview:** - **Architecture:** OPTForCausalLM (an autoregressive transformer for causal language modeling) - **Parameters:** ~98 million (97,838,592) - **Precision:** 32-bit - **Trained on:** kanishka/counterfactual_babylm_300_anans_new dataset - **Primary Task:** Text generation (causal language modeling) - **Evaluation Accuracy:** 0.4142 (on validation set) - **Validation Loss:** 3.3718 **Training Details:** - **Learning Rate:** 0.001 - **Batch Size:** 32 (train), 64 (eval) - **Epochs:** 20 - **Optimizer:** Adam (betas=(0.9,0.999), epsilon=1e-08) - **Scheduler:** Linear with 32,000 warmup steps - **Mixed Precision:** Native AMP **Intended Use:** - The model is designed for generating text in an autoregressive manner, suitable for research or applications involving language modeling, especially in counterfactual or controlled text generation scenarios. **Limitations:** - The model card does not specify detailed intended uses or limitations, so users should evaluate suitability for their specific use case. - No fine-tuned variants are currently listed. **Resources Required:** - **Disk Space:** 1 GB - **RAM:** 1 GB - **Recommended Instance:** g4dn.xlarge (micro) If you need more specific