Ajuste fino com o FunctionGemma

Acessar ai.google.dev Executar no Google Colab Executar no Kaggle Abrir na Vertex AI Ver código-fonte no GitHub

Este guia demonstra como ajustar o FunctionGemma para chamadas de ferramentas.

O FunctionGemma tem capacidade nativa de chamar ferramentas. Mas a verdadeira capacidade vem de duas habilidades distintas: o conhecimento mecânico de como usar uma ferramenta (sintaxe) e a capacidade cognitiva de interpretar por que e quando usá-la (intenção).

Os modelos, principalmente os menores, têm menos parâmetros disponíveis para manter a compreensão de intents complexas. Por isso, precisamos ajustar esses modelos.

Casos de uso comuns para ajuste fino da chamada de função:

  • Destilação de modelos: gerar dados de treinamento sintéticos com um modelo maior e ajustar um modelo menor para replicar o fluxo de trabalho específico com eficiência.
  • Processamento de esquemas não padrão: superação das dificuldades do modelo de base com estruturas de dados legadas, altamente complexas ou formatos proprietários não encontrados em dados públicos, como o processamento de ações móveis específicas do domínio.
  • Otimização do uso de contexto: "incorporação" de definições de ferramentas nos pesos do modelo. Isso permite usar descrições abreviadas nos comandos, liberando a janela de contexto para a conversa real.
  • Resolução da ambiguidade de seleção: direcionar o modelo para políticas empresariais específicas, como priorizar uma base de conhecimento interna em vez de um mecanismo de pesquisa externo.

Neste exemplo, vamos focar especificamente no gerenciamento da ambiguidade na seleção de ferramentas.

Configurar o ambiente de desenvolvimento

A primeira etapa é instalar as bibliotecas do Hugging Face, incluindo TRL e conjuntos de dados, para ajustar o modelo aberto, incluindo diferentes técnicas de RLHF e alinhamento.

# Install Pytorch & other libraries
%pip install torch tensorboard

# Install Hugging Face libraries
%pip install transformers datasets accelerate evaluate trl protobuf sentencepiece

# COMMENT IN: if you are running on a GPU that supports BF16 data type and flash attn, such as NVIDIA L4 or NVIDIA A100
#% pip install flash-attn

Observação: se você estiver usando uma GPU com arquitetura Ampere (como a NVIDIA L4) ou mais recente, poderá usar a atenção rápida. O Flash Attention é um método que acelera significativamente os cálculos e reduz o uso de memória de quadrático para linear no comprimento da sequência, acelerando o treinamento em até 3 vezes. Saiba mais em FlashAttention.

Antes de começar o treinamento, verifique se você aceitou os Termos de Uso da Gemma. Para aceitar a licença no Hugging Face, clique no botão Concordo e acesse o repositório na página do modelo em: http://huggingface.co/google/functiongemma-270m-it

Depois de aceitar a licença, você vai precisar de um token do Hugging Face válido para acessar o modelo. Se você estiver executando em um Google Colab, use seu token do Hugging Face com segurança usando os secrets do Colab. Caso contrário, defina o token diretamente no método login. Verifique se o token também tem acesso de gravação, já que você vai enviar o modelo para o Hub do Hugging Face após o ajuste refinado.

# Login into Hugging Face Hub
from huggingface_hub import login
login()

Você pode manter os resultados na máquina virtual local do Colab. No entanto, é altamente recomendável salvar os resultados intermediários no Google Drive. Isso garante que os resultados do treinamento estejam seguros e permite comparar e selecionar facilmente o melhor modelo.

Além disso, ajuste o diretório de pontos de verificação e a taxa de aprendizado.

from google.colab import drive

mount_google_drive = False
checkpoint_dir = "functiongemma-270m-it-simple-tool-calling"

if mount_google_drive:
    drive.mount('/content/drive')
    checkpoint_dir = f"/content/drive/MyDrive/{checkpoint_dir}"

print(f"Checkpoints will be saved to {checkpoint_dir}")

base_model = "google/functiongemma-270m-it"
learning_rate = 5e-5

Checkpoints will be saved to functiongemma-270m-it-simple-tool-calling

Preparar o conjunto de dados de ajuste refinado

Você vai usar o seguinte conjunto de dados de exemplo, que contém conversas de amostra que exigem uma escolha entre duas ferramentas: search_knowledge_base e search_google.

Conjunto de dados simples de chamada de ferramenta

Considere a consulta: "Quais são as práticas recomendadas para escrever uma função recursiva simples em Python?"

A ferramenta adequada depende totalmente da sua política específica. Embora um modelo genérico tenha como padrão search_google, um aplicativo empresarial geralmente precisa verificar search_knowledge_base primeiro.

Observação sobre a divisão de dados: para esta demonstração, você vai usar uma divisão de treinamento e teste de 50/50. Embora uma divisão de 80/20 seja padrão para fluxos de trabalho de produção, essa divisão igual é escolhida especificamente para destacar a melhoria de performance do modelo em dados não vistos.

import json
from datasets import Dataset
from transformers.utils import get_json_schema

# --- Tool Definitions ---
def search_knowledge_base(query: str) -> str:
    """
    Search internal company documents, policies and project data.

    Args:
        query: query string
    """
    return "Internal Result"

def search_google(query: str) -> str:
    """
    Search public information.

    Args:
        query: query string
    """
    return "Public Result"


TOOLS = [get_json_schema(search_knowledge_base), get_json_schema(search_google)]

DEFAULT_SYSTEM_MSG = "You are a model that can do function calling with the following functions"

def create_conversation(sample):
  return {
      "messages": [
          {"role": "developer", "content": DEFAULT_SYSTEM_MSG},
          {"role": "user", "content": sample["user_content"]},
          {"role": "assistant", "tool_calls": [{"type": "function", "function": {"name": sample["tool_name"], "arguments": json.loads(sample["tool_arguments"])} }]},
      ],
      "tools": TOOLS
  }

dataset = Dataset.from_list(simple_tool_calling)
# You can also load the dataset from Hugging Face Hub
# dataset = load_dataset("bebechien/SimpleToolCalling", split="train")

# Convert dataset to conversational format
dataset = dataset.map(create_conversation, remove_columns=dataset.features, batched=False)

# Split dataset into 50% training samples and 50% test samples
dataset = dataset.train_test_split(test_size=0.5, shuffle=True)

Map:   0%|          | 0/40 [00:00<?, ? examples/s]

Observação importante sobre a distribuição de conjuntos de dados

Ao usar o shuffle=False com seus próprios conjuntos de dados personalizados, verifique se os dados de origem estão pré-combinados. Se a distribuição for desconhecida ou classificada, use shuffle=True para garantir que o modelo aprenda uma representação equilibrada de todas as ferramentas durante o treinamento.

Ajustar o FunctionGemma usando TRL e SFTTrainer

Agora está tudo pronto para ajustar seu modelo. O SFTTrainer do Hugging Face TRL facilita a supervisão do ajuste de LLMs abertos. O SFTTrainer é uma subclasse do Trainer da biblioteca transformers e oferece suporte a todos os mesmos recursos.

O código a seguir carrega o modelo FunctionGemma e o tokenizador da Hugging Face.

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

# Load model and tokenizer
model = AutoModelForCausalLM.from_pretrained(
    base_model,
    dtype="auto",
    device_map="auto",
    attn_implementation="eager"
)
tokenizer = AutoTokenizer.from_pretrained(base_model)

print(f"Device: {model.device}")
print(f"DType: {model.dtype}")

# Print formatted user prompt
print("--- dataset input ---")
print(json.dumps(dataset["train"][0], indent=2))
debug_msg = tokenizer.apply_chat_template(dataset["train"][0]["messages"], tools=dataset["train"][0]["tools"], add_generation_prompt=False, tokenize=False)
print("--- Formatted prompt ---")
print(debug_msg)

Device: cuda:0
DType: torch.bfloat16
--- dataset input ---
{
  "messages": [
    {
      "content": "You are a model that can do function calling with the following functions",
      "role": "developer",
      "tool_calls": null
    },
    {
      "content": "What is the reimbursement limit for travel meals?",
      "role": "user",
      "tool_calls": null
    },
    {
      "content": null,
      "role": "assistant",
      "tool_calls": [
        {
          "function": {
            "arguments": {
              "query": "travel meal reimbursement limit policy"
            },
            "name": "search_knowledge_base"
          },
          "type": "function"
        }
      ]
    }
  ],
  "tools": [
    {
      "function": {
        "description": "Search internal company documents, policies and project data.",
        "name": "search_knowledge_base",
        "parameters": {
          "properties": {
            "query": {
              "description": "query string",
              "type": "string"
            }
          },
          "required": [
            "query"
          ],
          "type": "object"
        },
        "return": {
          "type": "string"
        }
      },
      "type": "function"
    },
    {
      "function": {
        "description": "Search public information.",
        "name": "search_google",
        "parameters": {
          "properties": {
            "query": {
              "description": "query string",
              "type": "string"
            }
          },
          "required": [
            "query"
          ],
          "type": "object"
        },
        "return": {
          "type": "string"
        }
      },
      "type": "function"
    }
  ]
}
--- Formatted prompt ---
<bos><start_of_turn>developer
You are a model that can do function calling with the following functions<start_function_declaration>declaration:search_knowledge_base{description:<escape>Search internal company documents, policies and project data.<escape>,parameters:{properties:{query:{description:<escape>query string<escape>,type:<escape>STRING<escape>} },required:[<escape>query<escape>],type:<escape>OBJECT<escape>} }<end_function_declaration><start_function_declaration>declaration:search_google{description:<escape>Search public information.<escape>,parameters:{properties:{query:{description:<escape>query string<escape>,type:<escape>STRING<escape>} },required:[<escape>query<escape>],type:<escape>OBJECT<escape>} }<end_function_declaration><end_of_turn>
<start_of_turn>user
What is the reimbursement limit for travel meals?<end_of_turn>
<start_of_turn>model
<start_function_call>call:search_knowledge_base{query:<escape>travel meal reimbursement limit policy<escape>}<end_function_call><start_function_response>

Antes do ajuste refinado

A saída abaixo mostra que os recursos prontos para uso podem não ser adequados para esse caso de uso.

def check_success_rate():
  success_count = 0
  for idx, item in enumerate(dataset['test']):
    messages = [
        item["messages"][0],
        item["messages"][1],
    ]

    inputs = tokenizer.apply_chat_template(messages, tools=TOOLS, add_generation_prompt=True, return_dict=True, return_tensors="pt")

    out = model.generate(**inputs.to(model.device), pad_token_id=tokenizer.eos_token_id, max_new_tokens=128)
    output = tokenizer.decode(out[0][len(inputs["input_ids"][0]) :], skip_special_tokens=False)

    print(f"{idx+1} Prompt: {item['messages'][1]['content']}")
    print(f"  Output: {output}")

    expected_tool = item['messages'][2]['tool_calls'][0]['function']['name']
    other_tool = "search_knowledge_base" if expected_tool == "search_google" else "search_google"

    if expected_tool in output and other_tool not in output:
      print("  `-> ✅ correct!")
      success_count += 1
    elif expected_tool not in output:
      print(f"  -> ❌ wrong (expected '{expected_tool}' missing)")
    else:
      if output.startswith(f"<start_function_call>call:{expected_tool}"):
        print(f"  -> ⚠️ tool is correct {expected_tool}, but other_tool exists in output")
      else:
        print(f"  -> ❌ wrong (hallucinated '{other_tool}')")

  print(f"Success : {success_count} / {len(dataset['test'])}")

check_success_rate()

1 Prompt: How do I access my paystubs on the ADP portal?
  Output: I cannot assist with accessing or retrieving paystubs or other company documents on the ADP portal. My current capabilities are limited to assisting with searching internal company documents and knowledge base queries.<end_of_turn>
  -> ❌ wrong (expected 'search_knowledge_base' missing)
2 Prompt: What is the syntax for Python list comprehensions?
  Output: I cannot assist with programming or providing programming syntax information. My current capabilities are focused on searching internal company documents and project data.<end_of_turn>
  -> ❌ wrong (expected 'search_google' missing)
3 Prompt: Where can I find the floor plan for Building B?
  Output: <start_function_call>call:search_knowledge_base{query:<escape>Floor plan for Building B<escape>}<end_function_call><start_function_call>call:search_google{query:<escape>Floor plan for Building B<escape>}<end_function_call><start_function_response>
  -> ⚠️ tool is correct search_knowledge_base, but other_tool exists in output
4 Prompt: Check the latest stock price for Apple.
  Output: <start_function_call>call:search_google{query:<escape>AAPL stock price<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
5 Prompt: What is the procedure for reporting a phishing email?
  Output: I cannot assist with providing instructions or procedures for reporting phishing emails. My capabilities are limited to assisting with specific search and document management functions.<end_of_turn>
  -> ❌ wrong (expected 'search_knowledge_base' missing)
6 Prompt: Show me examples of using the useEffect hook in React.
  Output: I am sorry, but I cannot assist with providing examples of using the `useEffect` hook in React. My current capabilities are focused on assisting with searching and retrieving internal company documents and project data using the specified tools. I cannot generate or explain code examples related to React hooks.<end_of_turn>
  -> ❌ wrong (expected 'search_google' missing)
7 Prompt: Who are the direct reports for the VP of Engineering?
  Output: <start_function_call>call:search_knowledge_base{query:<escape>VP of Engineering<escape>}<end_function_call><start_function_call>call:search_google{query:<escape>VP of Engineering<escape>}<end_function_call><start_function_response>
  -> ⚠️ tool is correct search_knowledge_base, but other_tool exists in output
8 Prompt: How do I list open ports on a Linux server?
  Output: I cannot assist with listing or querying open ports on Linux servers. My current capabilities are limited to assisting with searching internal company documents and project data using the specified tools.<end_of_turn>
  -> ❌ wrong (expected 'search_google' missing)
9 Prompt: What is our Slack message retention policy?
  Output: I cannot assist with finding or recommending company policies or terms of service regarding Slack messaging retention. My current capabilities are focused on searching internal company documents and project data.<end_of_turn>
  -> ❌ wrong (expected 'search_knowledge_base' missing)
10 Prompt: Compare the features of iPhone 15 vs Samsung S24.
  Output: I cannot assist with comparing device features or specifications. My current capabilities are focused on assisting with searching and managing company knowledge. I cannot browse or compare external product information.<end_of_turn>
  -> ❌ wrong (expected 'search_google' missing)
11 Prompt: I need the expense code for team building events.
  Output: I can certainly assist with searching for expense codes. Could you please specify the exact query you would like to use for searching the company expense database?<end_of_turn>
  -> ❌ wrong (expected 'search_knowledge_base' missing)
12 Prompt: Best practices for writing a Dockerfile for Node.js.
  Output: I am sorry, but I cannot assist with recommending best practices for writing Dockerfiles for Node.js. My current capabilities are focused on searching company documents and project data using specific tools. I cannot recommend or provide expert advice on software development best practices.<end_of_turn>
  -> ❌ wrong (expected 'search_google' missing)
13 Prompt: How do I request a new monitor setup?
  Output: I cannot assist with recommending or requesting hardware setup or configurations. My current capabilities are limited to assisting with searching company documents and project data using the specified tools. I cannot provide technical advice or recommendations for hardware setup.<end_of_turn>
  -> ❌ wrong (expected 'search_knowledge_base' missing)
14 Prompt: What is the difference between VLOOKUP and XLOOKUP in Google Sheets?
  Output: I cannot assist with comparing or contrasting Google Sheets and VLOOKUP/XLOOKUP in specific technical terms. My current capabilities are limited to assisting with specific data management functions like searching internal company documents and searching public knowledge base. I cannot provide expert comparisons or explanations of Google Sheets and VLOOKUP/XLOOKUP.<end_of_turn>
  -> ❌ wrong (expected 'search_google' missing)
15 Prompt: Find the onboarding checklist for new engineering hires.
  Output: <start_function_call>call:search_knowledge_base{query:<escape>onboarding checklist for new engineering hires<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
16 Prompt: What are the latest release notes for the OpenAI API?
  Output: <start_function_call>call:search_knowledge_base{query:<escape>OpenAI API release notes latest<escape>}<end_function_call><start_function_response>
  -> ❌ wrong (expected 'search_google' missing)
17 Prompt: Do we have preferred hotel partners in Paris?
  Output: I apologize, but I cannot assist with finding hotel partner information. My current capabilities are focused on searching company documents and project data using specific keywords. I cannot connect with or query business databases for hotel partnerships.<end_of_turn>
  -> ❌ wrong (expected 'search_knowledge_base' missing)
18 Prompt: How to undo the last git commit but keep the changes?
  Output: I cannot assist with managing or undoing Git repositories. My current capabilities are limited to assisting with searching and managing organizational knowledge bases. I cannot provide specific instructions or assistance with technical issues related to software development or Git operations.<end_of_turn>
  -> ❌ wrong (expected 'search_google' missing)
19 Prompt: What is the process for creating a new Jira project?
  Output: I cannot assist with creating or managing Jira project processes. My current capabilities are limited to assisting with searching company documents and project data using the specified tools.<end_of_turn>
  -> ❌ wrong (expected 'search_knowledge_base' missing)
20 Prompt: Tutorial on SQL window functions.
  Output: I cannot assist with tutorials or programming advice regarding SQL window functions. My capabilities are limited to assisting with searching company documents and knowledge base information.<end_of_turn>
  -> ❌ wrong (expected 'search_google' missing)
Success : 2 / 20

Treinamento

Antes de iniciar o treinamento, defina os hiperparâmetros que você quer usar em uma instância SFTConfig.

from trl import SFTConfig

torch_dtype = model.dtype

args = SFTConfig(
    output_dir=checkpoint_dir,              # directory to save and repository id
    max_length=512,                         # max sequence length for model and packing of the dataset
    packing=False,                          # Groups multiple samples in the dataset into a single sequence
    num_train_epochs=8,                     # number of training epochs
    per_device_train_batch_size=4,          # batch size per device during training
    gradient_checkpointing=False,           # Caching is incompatible with gradient checkpointing
    optim="adamw_torch_fused",              # use fused adamw optimizer
    logging_steps=1,                        # log every step
    #save_strategy="epoch",                  # save checkpoint every epoch
    eval_strategy="epoch",                  # evaluate checkpoint every epoch
    learning_rate=learning_rate,            # learning rate
    fp16=True if torch_dtype == torch.float16 else False,   # use float16 precision
    bf16=True if torch_dtype == torch.bfloat16 else False,  # use bfloat16 precision
    lr_scheduler_type="constant",            # use constant learning rate scheduler
    push_to_hub=True,                        # push model to hub
    report_to="tensorboard",                 # report metrics to tensorboard
)

Agora você tem todos os elementos básicos necessários para criar seu SFTTrainer e iniciar o treinamento do modelo.

from trl import SFTTrainer

# Create Trainer object
trainer = SFTTrainer(
    model=model,
    args=args,
    train_dataset=dataset['train'],
    eval_dataset=dataset['test'],
    processing_class=tokenizer,
)

Tokenizing train dataset:   0%|          | 0/20 [00:00<?, ? examples/s]
Truncating train dataset:   0%|          | 0/20 [00:00<?, ? examples/s]
Tokenizing eval dataset:   0%|          | 0/20 [00:00<?, ? examples/s]
Truncating eval dataset:   0%|          | 0/20 [00:00<?, ? examples/s]
The model is already on multiple devices. Skipping the move to device specified in `args`.

Comece o treinamento chamando o método train().

# Start training, the model will be automatically saved to the Hub and the output directory
trainer.train()

# Save the final model again to the Hugging Face Hub
trainer.save_model()

The tokenizer has new PAD/BOS/EOS tokens that differ from the model config and generation config. The model config and generation config were aligned accordingly, being updated with the tokenizer's values. Updated tokens: {'bos_token_id': 2, 'pad_token_id': 0}.

Para representar as perdas de treinamento e validação, normalmente você extrai esses valores do objeto TrainerState ou dos registros gerados durante o treinamento.

Bibliotecas como Matplotlib podem ser usadas para visualizar esses valores em etapas ou épocas de treinamento. O eixo x representa as etapas ou épocas de treinamento, e o eixo y representa os valores de perda correspondentes.

import matplotlib.pyplot as plt

# Access the log history
log_history = trainer.state.log_history

# Extract training / validation loss
train_losses = [log["loss"] for log in log_history if "loss" in log]
epoch_train = [log["epoch"] for log in log_history if "loss" in log]
eval_losses = [log["eval_loss"] for log in log_history if "eval_loss" in log]
epoch_eval = [log["epoch"] for log in log_history if "eval_loss" in log]

# Plot the training loss
plt.plot(epoch_train, train_losses, label="Training Loss")
plt.plot(epoch_eval, eval_losses, label="Validation Loss")
plt.xlabel("Epoch")
plt.ylabel("Loss")
plt.title("Training and Validation Loss per Epoch")
plt.legend()
plt.grid(True)
plt.show()

png

Teste de inferência do modelo

Depois do treinamento, você vai avaliar e testar o modelo. É possível carregar diferentes amostras do conjunto de dados de teste e avaliar o modelo nelas.

check_success_rate()

1 Prompt: How do I access my paystubs on the ADP portal?
  Output: <start_function_call>call:search_knowledge_base{query:<escape>paystubs API portal access codes<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
2 Prompt: What is the syntax for Python list comprehensions?
  Output: <start_function_call>call:search_google{query:<escape>Python list comprehensions syntax<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
3 Prompt: Where can I find the floor plan for Building B?
  Output: <start_function_call>call:search_knowledge_base{query:<escape>floor plan Building B floor plan<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
4 Prompt: Check the latest stock price for Apple.
  Output: <start_function_call>call:search_google{query:<escape>latest stock price Apple<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
5 Prompt: What is the procedure for reporting a phishing email?
  Output: <start_function_call>call:search_knowledge_base{query:<escape>phishing email procedure reporting policy<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
6 Prompt: Show me examples of using the useEffect hook in React.
  Output: <start_function_call>call:search_knowledge_base{query:<escape>useEffect hook examples React<escape>}<end_function_call><start_function_response>
  -> ❌ wrong (expected 'search_google' missing)
7 Prompt: Who are the direct reports for the VP of Engineering?
  Output: <start_function_call>call:search_knowledge_base{query:<escape>VP of Engineering direct reports<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
8 Prompt: How do I list open ports on a Linux server?
  Output: <start_function_call>call:search_google{query:<escape>open ports Linux server equivalents<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
9 Prompt: What is our Slack message retention policy?
  Output: <start_function_call>call:search_knowledge_base{query:<escape>slack message retention policy policy excerpt<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
10 Prompt: Compare the features of iPhone 15 vs Samsung S24.
  Output: <start_function_call>call:search_google{query:<escape>iPhone 15 vs Samsung S24 feature comparison<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
11 Prompt: I need the expense code for team building events.
  Output: <start_function_call>call:search_knowledge_base{query:<escape>expense code team building events<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
12 Prompt: Best practices for writing a Dockerfile for Node.js.
  Output: <start_function_call>call:search_knowledge_base{query:<escape>Docker file best practices Node.js<escape>}<end_function_call><start_function_response>
  -> ❌ wrong (expected 'search_google' missing)
13 Prompt: How do I request a new monitor setup?
  Output: <start_function_call>call:search_knowledge_base{query:<escape>new monitor setup request procedure<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
14 Prompt: What is the difference between VLOOKUP and XLOOKUP in Google Sheets?
  Output: <start_function_call>call:search_google{query:<escape>VLOOKUP vs XLOOKUP difference Google Sheets中<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
15 Prompt: Find the onboarding checklist for new engineering hires.
  Output: <start_function_call>call:search_knowledge_base{query:<escape>engineering hire onboarding checklist New hires.<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
16 Prompt: What are the latest release notes for the OpenAI API?
  Output: <start_function_call>call:search_google{query:<escape>latest OpenAI API release notes latest version<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
17 Prompt: Do we have preferred hotel partners in Paris?
  Output: <start_function_call>call:search_knowledge_base{query:<escape>preferred hotel partners in Paris<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
18 Prompt: How to undo the last git commit but keep the changes?
  Output: <start_function_call>call:search_knowledge_base{query:<escape>undo git commit last commit<escape>}<end_function_call><start_function_response>
  -> ❌ wrong (expected 'search_google' missing)
19 Prompt: What is the process for creating a new Jira project?
  Output: <start_function_call>call:search_knowledge_base{query:<escape>Jira project creation process<escape>}<end_function_call><start_function_response>
  `-> ✅ correct!
20 Prompt: Tutorial on SQL window functions.
  Output: <start_function_call>call:search_knowledge_base{query:<escape>SQL window functions tutorial<escape>}<end_function_call><start_function_response>
  -> ❌ wrong (expected 'search_google' missing)
Success : 16 / 20

Resumo e próximas etapas

Você aprendeu a ajustar o FunctionGemma para resolver a ambiguidade na seleção de ferramentas, um cenário em que um modelo precisa escolher entre ferramentas sobrepostas (por exemplo, pesquisa interna x externa) com base em políticas empresariais específicas. Ao usar a biblioteca TRL do Hugging Face e o SFTTrainer, o tutorial mostrou o processo de preparação de um conjunto de dados, configuração de hiperparâmetros e execução de um loop de ajuste refinado supervisionado.

Os resultados ilustram a diferença fundamental entre um modelo de base "capaz" e um modelo refinado "pronto para produção":

  • Antes do ajuste: o modelo de base tinha dificuldade em obedecer à política específica, muitas vezes não conseguindo chamar ferramentas ou escolhendo a errada, resultando em uma baixa taxa de sucesso (por exemplo, 2/20).
  • Após o ajuste refinado: depois de treinar por 8 épocas, o modelo aprendeu a distinguir corretamente entre consultas que exigem search_knowledge_base e search_google, melhorando a taxa de sucesso (por exemplo, 16/20).

Agora que você tem um modelo refinado, considere as seguintes etapas para avançar na produção:

  • Expanda o conjunto de dados: o conjunto de dados atual era uma pequena divisão sintética (50/50) usada para demonstração. Para um aplicativo empresarial robusto, organize um conjunto de dados maior e mais diversificado que cubra casos extremos e exceções raras de política.
  • Avaliação com RAG: integre o modelo ajustado a um pipeline de geração aumentada de recuperação (RAG) para verificar se as chamadas de ferramenta search_knowledge_base realmente recuperam documentos relevantes e resultam em respostas finais precisas.

Confira os seguintes documentos: