Create README.md

README.md

@@ -0,0 +1,81 @@
+---
+license: apache-2.0
+language:
+- en
+base_model:
+- PygmalionAI/Eleusis-12B
+pipeline_tag: text-generation
+---
+
+This is an ONNX optimized version of [Eleusis-12B](https://huggingface.co/PygmalionAI/Eleusis-12B).
+For a more comprehensive info about the model's capabilities, please visit the original model's repo.
+
+## Inference
+### Requirements
+If you're on a CPU-only machine:
+
+```sh
+pip install onnxruntime
+```
+
+If you have an NVIDIA GPU available:
+```sh
+pip uninstall onnxruntime -y
+pip install onnxruntime-gpu
+```
+
+Make sure you have installed [CUDA Toolkit](https://developer.nvidia.com/cuda-12-4-0-download-archive) and [cuDNN](https://developer.nvidia.com/cudnn)
+
+```sh
+import onnxruntime as ort
+from transformers import AutoTokenizer
+import numpy as np
+import argparse
+
+def generate_text(prompt, num_tokens, model_path, tokenizer_path):
+    tokenizer = AutoTokenizer.from_pretrained(tokenizer_path)
+    providers = ['CUDAExecutionProvider', 'CPUExecutionProvider']
+    session = ort.InferenceSession(model_path, providers=providers)
+
+    input_ids = tokenizer(prompt, return_tensors="np").input_ids
+
+    for _ in range(num_tokens):
+        # Create attention mask and position ids
+        attention_mask = np.ones_like(input_ids)
+        position_ids = np.arange(input_ids.shape[1])[None, :]
+
+        outputs = session.run(
+            output_names=['logits'],
+            input_feed={
+                'input_ids': input_ids,
+                'attention_mask': attention_mask,
+                'position_ids': position_ids
+            }
+        )
+
+        next_token = np.argmax(outputs[0][0, -1, :])
+
+        input_ids = np.concatenate([input_ids, [[next_token]]], axis=1)
+
+    return tokenizer.decode(input_ids[0], skip_special_tokens=True)
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description='Generate text using ONNX model')
+    parser.add_argument('prompt', type=str, help='Input prompt for generation')
+    parser.add_argument('num_tokens', type=int, help='Number of tokens to generate')
+    parser.add_argument('--model_path', type=str, default='model.onnx', 
+                        help='Path to ONNX model file')
+    parser.add_argument('--tokenizer_path', type=str, default='tokenizer',
+                        help='Path to tokenizer directory')
+
+    args = parser.parse_args()
+
+    result = generate_text(args.prompt, args.num_tokens, args.model_path, args.tokenizer_path)
+    print(result)
+```
+
+```sh
+python onnx_inference.py "Once upon a time" 512 --model_path /path/to/model.onnx --tokenizer_path /path/to/model/dir
+```
+
+This is an example script, and not properly optimized.