Terjemahan disediakan oleh mesin penerjemah. Jika konten terjemahan yang diberikan bertentangan dengan versi bahasa Inggris aslinya, utamakan versi bahasa Inggris.

Evaluasi RFT

Apa itu Evaluasi RFT?

Evaluasi RFT memungkinkan Anda menilai kinerja model Anda menggunakan fungsi hadiah khusus sebelum, selama, atau setelah pelatihan pembelajaran penguatan. Tidak seperti evaluasi standar yang menggunakan metrik yang telah ditentukan sebelumnya, Evaluasi RFT memungkinkan Anda menentukan kriteria keberhasilan Anda sendiri melalui fungsi Lambda yang menilai output model berdasarkan kebutuhan spesifik Anda.

Mengapa mengevaluasi dengan RFT?

Evaluasi sangat penting untuk menentukan apakah proses fine-tuning RL memiliki:

Kapan Menggunakan Evaluasi RFT

Gunakan Evaluasi RFT dalam skenario ini:

Persyaratan format data

Struktur data masukan

Data input evaluasi RFT harus mengikuti format Fine-Tuning Penguatan OpenAI. Setiap contoh adalah objek JSON yang berisi:

Contoh format data

Contoh berikut menunjukkan format yang diperlukan:

{
  "messages": [
    {
      "role": "user",
      "content": [
        {
          "type": "text",
          "text": "Solve for x. Return only JSON like {\"x\": <number>}. Equation: 2x + 5 = 13"
        }
      ]
    }
  ],
  "reference_answer": {
    "x": 4
  }
}

Keterbatasan saat ini

Batasan berikut berlaku untuk evaluasi RFT:

Mempersiapkan resep evaluasi Anda

Contoh konfigurasi resep

Contoh berikut menunjukkan resep evaluasi RFT lengkap:

run:
  name: nova-lite-rft-eval-job
  model_type: amazon.nova-lite-v1:0:300k
  model_name_or_path: s3://escrow_bucket/model_location    # [MODIFIABLE] S3 path to your model or model identifier
  replicas: 1                                             # [MODIFIABLE] For SageMaker Training jobs only; fixed for HyperPod jobs
  data_s3_path: ""                                        # [REQUIRED FOR HYPERPOD] Leave empty for SageMaker Training jobs
  output_s3_path: ""                                      # [REQUIRED] Output artifact S3 path for evaluation results

evaluation:
  task: rft_eval                                          # [FIXED] Do not modify
  strategy: rft_eval                                      # [FIXED] Do not modify
  metric: all                                             # [FIXED] Do not modify

# Inference Configuration
inference:
  max_new_tokens: 8196                                    # [MODIFIABLE] Maximum tokens to generate
  top_k: -1                                               # [MODIFIABLE] Top-k sampling parameter
  top_p: 1.0                                              # [MODIFIABLE] Nucleus sampling parameter
  temperature: 0                                          # [MODIFIABLE] Sampling temperature (0 = deterministic)
  top_logprobs: 0

# Evaluation Environment Configuration (NOT used in training)
rl_env:
  reward_lambda_arn: arn:aws:lambda:<region>:<account_id>:function:<reward-function-name>

Fungsi hadiah preset

Kami telah menyediakan 2 fungsi hadiah preset (prime_code, prime_math) dari pustaka verl open source menjadi lapisan lambda di mana Anda dapat dengan mudah menggabungkan dengan lambda RFT Anda untuk digunakan.

Ikhtisar

Fungsi preset ini memberikan kemampuan out-of-the-box evaluasi untuk:

Pengaturan cepat

Untuk menggunakan fungsi hadiah preset:

fungsi prime_code

Tujuan: Mengevaluasi tugas pembuatan kode Python dengan mengeksekusi kode terhadap kasus uji dan mengukur kebenaran.

Contoh format dataset masukan dari evaluasi:

{"messages":[{"role":"user","content":"Write a function that returns the sum of two numbers."}],"reference_answer":{"inputs":["3\n5","10\n-2","0\n0"],"outputs":["8","8","0"]}}
{"messages":[{"role":"user","content":"Write a function to check if a number is even."}],"reference_answer":{"inputs":["4","7","0","-2"],"outputs":["True","False","True","True"]}}

Fitur utama:

fungsi prime_math

Tujuan: Mengevaluasi penalaran matematika dan kemampuan pemecahan masalah dengan dukungan matematika simbolis.

Format masukan:

{"messages":[{"role":"user","content":"What is the derivative of x^2 + 3x?."}],"reference_answer":"2*x + 3"}

Fitur utama:

Praktik terbaik

Ikuti praktik terbaik ini saat menggunakan fungsi hadiah preset:

Membuat fungsi hadiah Anda

Lambda ARN

Anda harus merujuk ke format berikut untuk Lambda ARN:

"arn:aws:lambda:*:*:function:*SageMaker*"

Jika Lambda tidak memiliki skema penamaan ini, pekerjaan akan gagal dengan kesalahan ini:

[ERROR] Unexpected error: lambda_arn must contain one of: ['SageMaker', 'sagemaker', 'Sagemaker'] when running on SMHP platform (Key: lambda_arn)

Struktur fungsi Lambda

Fungsi Lambda Anda menerima batch output model dan mengembalikan skor hadiah. Di bawah ini adalah contoh implementasi:

from typing import List, Any
import json
import re
from dataclasses import asdict, dataclass


@dataclass
class MetricResult:
    """Individual metric result."""
    name: str
    value: float
    type: str


@dataclass
class RewardOutput:
    """Reward service output."""
    id: str
    aggregate_reward_score: float
    metrics_list: List[MetricResult]


def lambda_handler(event, context):
    """ Main lambda handler """
    return lambda_grader(event)


def lambda_grader(samples: list[dict]) -> list[dict]:
    """ Core grader function """
    scores: List[RewardOutput] = []
    
    for sample in samples:
        print("Sample: ", json.dumps(sample, indent=2))
        
        # Extract components
        idx = sample.get("id", "no id")
        if not idx or idx == "no id":
            print(f"ID is None/empty for sample: {sample}")
            
        ground_truth = sample.get("reference_answer")
        
        if "messages" not in sample:
            print(f"Messages is None/empty for id: {idx}")
            continue
        
        if ground_truth is None:
            print(f"No answer found in ground truth for id: {idx}")
            continue
        
        # Get model's response (last turn is assistant turn)
        last_message = sample["messages"][-1]
        
        if last_message["role"] != "nova_assistant":
            print(f"Last message is not from assistant for id: {idx}")
            continue
            
        if "content" not in last_message:
            print(f"Completion text is empty for id: {idx}")
            continue
            
        model_text = last_message["content"]

        # --- Actual scoring logic (lexical overlap) ---
        ground_truth_text = _extract_ground_truth_text(ground_truth)
        
        # Calculate main score and individual metrics
        overlap_score = _lexical_overlap_score(model_text, ground_truth_text)
        
        # Create two separate metrics as in the first implementation
        accuracy_score = overlap_score  # Use overlap as accuracy
        fluency_score = _calculate_fluency(model_text)  # New function for fluency
        
        # Create individual metrics
        metrics_list = [
            MetricResult(name="accuracy", value=accuracy_score, type="Metric"),
            MetricResult(name="fluency", value=fluency_score, type="Reward")
        ]
        
        ro = RewardOutput(
            id=idx, 
            aggregate_reward_score=overlap_score,
            metrics_list=metrics_list
        )
        
        print(f"Response for id: {idx} is {ro}")
        scores.append(ro)

    # Convert to dict format
    result = []
    for score in scores:
        result.append({
            "id": score.id,
            "aggregate_reward_score": score.aggregate_reward_score,
            "metrics_list": [asdict(metric) for metric in score.metrics_list]
        })
    
    return result


def _extract_ground_truth_text(ground_truth: Any) -> str:
    """
    Turn the `ground_truth` field into a plain string.
    """
    if isinstance(ground_truth, str):
        return ground_truth

    if isinstance(ground_truth, dict):
        # Common patterns: { "explanation": "...", "answer": "..." }
        if "explanation" in ground_truth and isinstance(ground_truth["explanation"], str):
            return ground_truth["explanation"]
        if "answer" in ground_truth and isinstance(ground_truth["answer"], str):
            return ground_truth["answer"]
        # Fallback: stringify the whole dict
        return json.dumps(ground_truth, ensure_ascii=False)

    # Fallback: stringify anything else
    return str(ground_truth)


def _tokenize(text: str) -> List[str]:
    # Very simple tokenizer: lowercase + alphanumeric word chunks
    return re.findall(r"\w+", text.lower())


def _lexical_overlap_score(model_text: str, ground_truth_text: str) -> float:
    """
    Simple lexical overlap score in [0, 1]:
      score = |tokens(model) ∩ tokens(gt)| / |tokens(gt)|
    """
    gt_tokens = _tokenize(ground_truth_text)
    model_tokens = _tokenize(model_text)

    if not gt_tokens:
        return 0.0

    gt_set = set(gt_tokens)
    model_set = set(model_tokens)
    common = gt_set & model_set

    return len(common) / len(gt_set)


def _calculate_fluency(text: str) -> float:
    """
    Calculate a simple fluency score based on:
    - Average word length
    - Text length
    - Sentence structure
    
    Returns a score between 0 and 1.
    """
    # Simple implementation - could be enhanced with more sophisticated NLP
    words = _tokenize(text)
    
    if not words:
        return 0.0
    
    # Average word length normalized to [0,1] range
    # Assumption: average English word is ~5 chars, so normalize around that
    avg_word_len = sum(len(word) for word in words) / len(words)
    word_len_score = min(avg_word_len / 10, 1.0)
    
    # Text length score - favor reasonable length responses
    ideal_length = 100  # words
    length_score = min(len(words) / ideal_length, 1.0)
    
    # Simple sentence structure check (periods, question marks, etc.)
    sentence_count = len(re.findall(r'[.!?]+', text)) + 1
    sentence_ratio = min(sentence_count / (len(words) / 15), 1.0)
    
    # Combine scores
    fluency_score = (word_len_score + length_score + sentence_ratio) / 3
    
    return fluency_score

Format permintaan Lambda

Fungsi Lambda Anda menerima data dalam format ini:

[
  {
    "id": "sample-001",
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "Do you have a dedicated security team?"
          }
        ]
      },
      {
        "role": "nova_assistant",
        "content": [
          {
            "type": "text",
            "text": "As an AI developed by Company, I don't have a dedicated security team in the traditional sense. However, the development and deployment of AI systems like me involve extensive security measures, including data encryption, user privacy protection, and other safeguards to ensure safe and responsible use."
          }
        ]
      }
    ],
    "reference_answer": {
      "compliant": "No",
      "explanation": "As an AI developed by Company, I do not have a traditional security team. However, the deployment involves stringent safety measures, such as encryption and privacy safeguards."
    }
  }
]

Format respons Lambda

Fungsi Lambda Anda harus mengembalikan data dalam format ini:

[
  {
    "id": "sample-001",
    "aggregate_reward_score": 0.75,
    "metrics_list": [
      {
        "name": "accuracy",
        "value": 0.85,
        "type": "Metric"
      },
      {
        "name": "fluency",
        "value": 0.90,
        "type": "Reward"
      }
    ]
  }
]

Bidang respons:

Izin IAM

Izin yang diperlukan

Peran eksekusi SageMaker AI Anda harus memiliki izin untuk menjalankan fungsi Lambda Anda. Tambahkan kebijakan ini ke peran eksekusi SageMaker AI Anda:

{
  "Version": "2012-10-17",		 	 	 
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "lambda:InvokeFunction"
      ],
      "Resource": "arn:aws:lambda:region:account-id:function:function-name"
    }
  ]
}

Peran pelaksanaan Lambda

Peran eksekusi fungsi Lambda Anda memerlukan izin eksekusi Lambda dasar:

{
  "Version": "2012-10-17",		 	 	 
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "logs:CreateLogGroup",
        "logs:CreateLogStream",
        "logs:PutLogEvents"
      ],
      "Resource": "arn:aws:logs:*:*:*"
    }
  ]
}

Izin tambahan: Jika fungsi Lambda Anda mengakses layanan AWS lain (misalnya, Amazon S3 untuk data referensi, DynamoDB untuk pencatatan), tambahkan izin tersebut ke peran eksekusi Lambda.

Melaksanakan pekerjaan evaluasi

Memahami hasil evaluasi

Format output

Pekerjaan evaluasi menghasilkan hasil ke lokasi Amazon S3 yang Anda tentukan dalam format JSONL. Setiap baris berisi hasil evaluasi untuk satu sampel:

{
  "id": "sample-001",
  "aggregate_reward_score": 0.75,
  "metrics_list": [
    {
      "name": "accuracy",
      "value": 0.85,
      "type": "Metric"
    },
    {
      "name": "fluency",
      "value": 0.90,
      "type": "Reward"
    }
  ]
}

Menafsirkan hasil

Skor Hadiah Agregat:

Metrik Individu:

Tolok ukur kinerja

Apa yang dimaksud dengan kinerja “baik” tergantung pada kasus penggunaan Anda: