README.md

metadata

base_model:
  - Qwen/Qwen2.5-7B-Instruct
language:
  - en
  - zh
license: mit
pipeline_tag: question-answering
library_name: transformers
tags:
  - biology
  - finance
  - text-generation-inference

HierSearch: A Hierarchical Enterprise Deep Search Framework Integrating Local and Web Searches

Model Information

We release the agent model used in HierSearch: A Hierarchical Enterprise Deep Search Framework Integrating Local and Web Searches.

Useful links: 📝 Paper (arXiv) • 🤗 Paper (Hugging Face) • 🧩 Github

1. We explore the deep search framework in multi-knowledge-source scenarios and propose a hierarchical agentic paradigm and train with HRL;
2. We notice drawbacks of the naive information transmission among deep search agents and developed a knowledge refiner suitable for multi-knowledge-source scenarios;
3. Our proposed approach for reliable and effective deep search across multiple knowledge sources outperforms existing baselines the flat-RL solution in various domains.

🌹 If you use this model, please ✨star our GitHub repository or upvote our paper to support us. Your star means a lot!

Sample Usage

You can load and use this model directly with the Hugging Face transformers library for basic text generation or question-answering inference. For the full HierSearch framework capabilities, please refer to the official GitHub repository.

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "zstanjj/HierSearch-Planner-Agent" # This model represents the Planner Agent.
# Other agent models include "zstanjj/HierSearch-Local-Agent" or "zstanjj/HierSearch-Web-Agent".

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16, # Or torch.float16 depending on your hardware
    device_map="auto" # Or specify your device, e.g., "cuda:0"
)

# Example for a question-answering interaction with the Planner Agent
messages = [
    {"role": "user", "content": "Explain the concept of Hierarchical Reinforcement Learning as applied in this paper."},
]

# Apply chat template and tokenize inputs
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)

model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# Generate response
generated_ids = model.generate(
    model_inputs.input_ids,
    max_new_tokens=1024, # Adjust max_new_tokens as needed for detailed answers
    temperature=0.7,    # Adjust generation parameters for diversity
    do_sample=True,
    eos_token_id=tokenizer.eos_token_id, # Ensure generation stops at EOS token
    pad_token_id=tokenizer.pad_token_id # Set pad_token_id for proper generation
)

# Decode and print the output
decoded_output = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(decoded_output)