ragflow/common/token_utils.py

#
#  Copyright 2025 The InfiniFlow Authors. All Rights Reserved.
#
#  Licensed under the Apache License, Version 2.0 (the "License");
#  you may not use this file except in compliance with the License.
#  You may obtain a copy of the License at
#
#      http://www.apache.org/licenses/LICENSE-2.0
#
#  Unless required by applicable law or agreed to in writing, software
#  distributed under the License is distributed on an "AS IS" BASIS,
#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#  See the License for the specific language governing permissions and
#  limitations under the License.
#

import hashlib
import os
import shutil
import tiktoken

from common.file_utils import get_project_base_directory


def _ensure_tiktoken_cache() -> str:
    cache_dir = get_project_base_directory()
    os.environ["TIKTOKEN_CACHE_DIR"] = cache_dir

    bundled_encoding_path = get_project_base_directory("ragflow_deps", "cl100k_base.tiktoken")
    encoding_url = "https://openaipublic.blob.core.windows.net/encodings/cl100k_base.tiktoken"
    cached_encoding_path = os.path.join(cache_dir, hashlib.sha1(encoding_url.encode()).hexdigest())

    if os.path.exists(bundled_encoding_path) and not os.path.exists(cached_encoding_path):
        shutil.copyfile(bundled_encoding_path, cached_encoding_path)

    return cache_dir


tiktoken_cache_dir = _ensure_tiktoken_cache()
os.environ["TIKTOKEN_CACHE_DIR"] = tiktoken_cache_dir
# encoder = tiktoken.encoding_for_model("gpt-3.5-turbo")
encoder = tiktoken.get_encoding("cl100k_base")


def num_tokens_from_string(string: str) -> int:
    """Returns the number of tokens in a text string."""
    try:
        code_list = encoder.encode(string)
        return len(code_list)
    except Exception:
        return 0

def total_token_count_from_response(resp):
    """
    Extract token count from LLM response in various formats.

    Handles None responses and different response structures from various LLM providers.
    Returns 0 if token count cannot be determined.
    """
    if resp is None:
        return 0

    try:
        if hasattr(resp, "usage") and hasattr(resp.usage, "total_tokens"):
            return resp.usage.total_tokens
    except Exception:
        pass

    try:
        if hasattr(resp, "usage_metadata") and hasattr(resp.usage_metadata, "total_tokens"):
            return resp.usage_metadata.total_tokens
    except Exception:
        pass

    try:
        if hasattr(resp, "meta") and hasattr(resp.meta, "billed_units") and hasattr(resp.meta.billed_units, "input_tokens"):
            return resp.meta.billed_units.input_tokens
    except Exception:
        pass

    if isinstance(resp, dict) and 'usage' in resp and 'total_tokens' in resp['usage']:
        try:
            return resp["usage"]["total_tokens"]
        except Exception:
            pass

    if isinstance(resp, dict) and 'usage' in resp and 'input_tokens' in resp['usage'] and 'output_tokens' in resp['usage']:
        try:
            return resp["usage"]["input_tokens"] + resp["usage"]["output_tokens"]
        except Exception:
            pass

    if isinstance(resp, dict) and 'meta' in resp and 'tokens' in resp['meta'] and 'input_tokens' in resp['meta']['tokens'] and 'output_tokens' in resp['meta']['tokens']:
        try:
            return resp["meta"]["tokens"]["input_tokens"] + resp["meta"]["tokens"]["output_tokens"]
        except Exception:
            pass
    return 0


def truncate(string: str, max_len: int) -> str:
    """Returns truncated text if the length of text exceed max_len."""
    return encoder.decode(encoder.encode(string)[:max_len])