Update README.md

README.md

@@ -1,3 +1,96 @@
 ---
 license: cc-by-nc-sa-4.0
+language:
+- 'no'
 ---
+
+
+# Model Card
+
+NorGPT-369M-NO-QNLI-peft is trained on top of [NorGPT-369M](https://huggingface.co/NorGLM/NorGPT-369M) model on [NO-QNLI](https://huggingface.co/datasets/NorGLM/NO-QNLI) dataset.
+
+Data format:
+```
+input: {premise}[SEP]{hypothesis}
+label: {entailment, not_entailment} -> {1,0}
+```
+
+## Run the Model
+```python
+from peft import PeftModel, PeftConfig
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+torch_device = "cuda" if torch.cuda.is_available() else "cpu"
+
+source_model_id = "NbAiLab/nb-gpt-j-6B"
+peft_model_id = "NorGLM/NorGPT-369M-NO-QNLI-peft"
+
+config = PeftConfig.from_pretrained(peft_model_id)
+model = AutoModelForCausalLM.from_pretrained(source_model_id, device_map='balanced')
+
+tokenizer_max_len = 2048
+tokenizer_config = {'pretrained_model_name_or_path': source_model_id,
+                            'max_len': tokenizer_max_len}
+tokenizer = tokenizer = AutoTokenizer.from_pretrained(**tokenizer_config)
+tokenizer.pad_token = tokenizer.eos_token
+
+model = PeftModel.from_pretrained(model, peft_model_id)
+```
+
+## Inference Example
+Load the model to evaluate on the validation set:
+```python
+
+def getDataSetFromFiles(df):
+    # convert dataset    
+    df["text"] = df[["premise", "hypothesis"]].apply(lambda x: " [SEP] ".join(x.astype(str)), axis =1)
+    df = df.drop(["idx", "premise", "hypothesis"], axis=1)
+    #df['label'] = df['label'].replace({1:'contradiction', -1:'entailment', 0:'neutral'})
+    df["label"] = df.label.map({"not_entailment": 0, "entailment": 1})
+    return Dataset.from_pandas(df)
+
+print("--LOADING EVAL DATAS---")
+eval_data = load_dataset("NorGLM/NO-BoolQ", data_files="val.jsonl")
+eval_data = getDataSetFromFiles(eval_data["train"].to_pandas())
+
+print("--MAKING PREDICTIONS---")
+model.eval()
+
+y_true = []
+y_pred = []
+count = 0
+
+for data in eval_data:
+    count = count + 1
+    if count % 100 == 0:
+        print(count)
+    inputs = tokenizer(data['text'], return_tensors="pt").to(torch_device)
+    
+    with torch.no_grad():
+        logits = model(**inputs).logits
+        #print(logits)
+
+    predicted_class_id = logits.argmax().item()
+
+    y_true.append(data['label'])
+    y_pred.append(predicted_class_id)
+
+print(y_pred)
+
+print(f"Lenght of true_values: {len(y_true)}")   
+print(f"Lenght of predicted_values: {len(y_pred)}")    
+
+y_true = np.array(y_true)
+y_pred = np.array(y_pred)
+
+F_score = f1_score(y_true, y_pred, average="macro")
+print(f"F1 score: {F_score}")
+
+accuracy = accuracy_score(y_true, y_pred)
+print(f"Accuracy: {accuracy}")
+
+```
+
+## Note
+More training details will be released soon!