Create README.md

README.md

@@ -0,0 +1,59 @@
+---
+language:
+- it
+- en
+license: llama3
+library_name: transformers
+base_model: DeepMount00/Llama-3-8b-Ita
+datasets:
+- DeepMount00/llm_ita_ultra
+---
+# QuantFactory/Llama-3-8b-Ita-GGUF
+This is quantized version of [DeepMount00/Llama-3-8b-Ita](https://huggingface.co/DeepMount00/Llama-3-8b-Ita) created using llama.cpp
+
+## Model Description
+- **Base Model:** [Meta-Llama-3-8B](https://huggingface.co/meta-llama/Meta-Llama-3-8B)
+- **Specialization:** Italian Language
+
+## Evaluation
+
+For a detailed comparison of model performance, check out the [Leaderboard for Italian Language Models](https://huggingface.co/spaces/FinancialSupport/open_ita_llm_leaderboard).
+
+Here's a breakdown of the performance metrics:
+
+| Metric                      | hellaswag_it acc_norm | arc_it acc_norm | m_mmlu_it 5-shot acc | Average |
+|:----------------------------|:----------------------|:----------------|:---------------------|:--------|
+| **Accuracy Normalized**     | 0.6518                | 0.5441        | 0.5729              | 0.5896  |
+
+---
+
+## How to Use
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+MODEL_NAME = "DeepMount00/Llama-3-8b-Ita"
+
+model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, torch_dtype=torch.bfloat16).eval()
+model.to(device)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+
+def generate_answer(prompt):
+    messages = [
+        {"role": "user", "content": prompt},
+    ]
+    model_inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to(device)
+    generated_ids = model.generate(model_inputs, max_new_tokens=200, do_sample=True,
+                                          temperature=0.001)
+    decoded = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
+    return decoded[0]
+
+prompt = "Come si apre un file json in python?"
+answer = generate_answer(prompt)
+print(answer)
+```
+---
+## Developer
+[Michele Montebovi]