Create README.md

README.md

@@ -0,0 +1,95 @@
+---
+base_model:
+- google/gemma-2-2b-it
+tags:
+- text-generation-inference
+- transformers
+- unsloth
+- gemma2
+- trl
+license: gemma
+language:
+- en
+- fi
+- sv
+---
+
+This example utilizes the [European AI Act regulation text](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689) as training data in three languages: 
+English, Finnish, and Swedish. The dataset comprises 9,175 data points for training and 2,456 for evaluation. 
+
+Python libraries needed:
+
+```python
+pip install -U transformers
+pip install torch torchvision torchaudio
+pip install 'accelerate>=0.26.0'
+```
+
+The training arguments used are as follows:
+
+```python
+training_args = TrainingArguments(
+    per_device_train_batch_size=32,  
+    gradient_accumulation_steps=32,  
+    warmup_steps=10,
+    max_steps=200,
+    learning_rate=2e-5,  
+    fp16=not is_bfloat16_supported(),
+    bf16=is_bfloat16_supported(),
+    logging_steps=1,
+    optim="adamw_8bit",
+    weight_decay=0.01,
+    lr_scheduler_type="linear",  
+    seed=3407,
+    output_dir=output_dir,
+    report_to="none",
+    eval_strategy="steps",
+    eval_steps=20,
+    load_best_model_at_end=True, 
+    metric_for_best_model="eval_loss",
+    greater_is_better=False,
+    save_total_limit=2, 
+)
+```
+
+The prediction is made using the standard Gemma:
+
+```python
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import transformers
+import torch
+
+model_id = "mlconvexai/gemma-2-2b-it-finetuned-EU-Act"
+dtype = torch.bfloat16
+
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+model = AutoModelForCausalLM.from_pretrained(
+    model_id,
+    device_map="auto",
+    torch_dtype=dtype,)
+
+chat = [
+    { "role": "user", "content": "Mikä on EU:n tekoälyasetus?" },
+]
+prompt = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
+inputs = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt")
+outputs = model.generate(
+    input_ids=inputs.to(model.device), 
+    max_new_tokens=1024,
+    repetition_penalty=1.1,
+    no_repeat_ngram_size=4,
+)
+print(tokenizer.decode(outputs[0]))
+```
+
+More detailed information about fine-tuning can be found on [Medium](https://medium.com/@timo.au.laine/eu-ai-act-fine-tune-multilingual-local-llm-2c0657cc47f8).
+
+# Uploaded  model
+
+- **Developed by:** mlconvexai
+- **License:** Gemma
+- **Finetuned from model :** google/gemma-2-2b-it
+
+This gemma2 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
+
+[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)