ضبط Gemma لاستدعاء الدوال

import os
from kaggle_secrets import UserSecretsClient
user_secrets = UserSecretsClient()
os.environ['HF_TOKEN'] = user_secrets.get_secret("HF_TOKEN")

!pip install transformers==4.46.1 datasets==3.1.0 trl==0.12.0 peft==0.13.2 accelerate==0.34.0 torch~=2.5.0 torch_xla[tpu]~=2.5.0 tpu-info

from transformers import AutoTokenizer, AutoModelForCausalLM
model_name = "google/gemma-2-2b-it"
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16)
model.config.use_cache = False
tokenizer = AutoTokenizer.from_pretrained(model_name)
tokenizer.padding_side = "right"

from datasets import load_dataset
dataset = load_dataset("lilacai/glaive-function-calling-v2-sharegpt", split="train[:15%]")

chat_template = "{{ bos_token }}{% if messages[0]['from'] == 'system' %}{{'<start_of_turn>user\n' + messages[0]['value'] | trim + ' ' + messages[1]['value'] | trim + '<end_of_turn>\n'}}{% set messages = messages[2:] %}{% endif %}{% for message in messages %}{% if message['from'] == 'human' %}{{'<start_of_turn>user\n' + message['value'] | trim + '<end_of_turn>\n'}}{% elif message['from'] == 'gpt' %}{{'<start_of_turn>model\n' + message['value'] | trim + '<end_of_turn>\n' }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '<start_of_turn>model\n' }}{% endif %}"
tokenizer.chat_template = chat_template

from peft import LoraConfig
peft_config = LoraConfig(lora_alpha=16, lora_dropout=0, r=16, bias="none", task_type="CAUSAL_LM", target_modules=["q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj"])

from trl import SFTTrainer, SFTConfig
training_arguments = SFTConfig(output_dir="./results", max_steps=100, per_device_train_batch_size=32, optim="adafactor", learning_rate=0.0002, bf16=True, max_seq_length=1024, dataset_text_field="text", packing=True, logging_steps=1, seed=42)

trainer = SFTTrainer(model=model, train_dataset=dataset, peft_config=peft_config, args=training_arguments)
trainer.train()

trainer.model.to('cpu').save_pretrained("gemma-func-ft")

base_model = AutoModelForCausalLM.from_pretrained(model_name, low_cpu_mem_usage=True, torch_dtype=torch.bfloat16)
model = PeftModel.from_pretrained(base_model, "gemma-func-ft")
model = model.merge_and_unload()
input_text = "<start_of_turn>user\nأنت مساعد مفيد..."
input_ids = tokenizer(input_text, return_tensors="pt").to("cpu")
outputs = model.generate(**input_ids, max_new_tokens=512)
print(tokenizer.decode(outputs[0]))