muradil211/AetherSearch

1---2library_name: transformers3pipeline_tag: text-generation4tags:5  - aethersearch6  - agentic-rl7  - search-augmented-generation8  - qwen29  - sft10  - dpo11  - reinforcement-learning12language:13  - en14---15 16<div align="center">17<img src="assets/aethersearch-mark.svg" alt="AetherSearch monogram" width="128">18</div>19 20# AetherSearch21 22AetherSearch is a search-augmented language model release trained through a23multi-stage post-training pipeline with SFT, DPO, and reinforcement learning.24 25The released weights are provided in Hugging Face Transformers format and can be26loaded with `AutoModelForCausalLM` and `AutoTokenizer`.27 28## Files29 30- `model.safetensors`: model weights.31- `config.json` and `generation_config.json`: Transformers configuration.32- `tokenizer.json`, `tokenizer_config.json`, `vocab.json`, `merges.txt`,33  `added_tokens.json`, and `special_tokens_map.json`: tokenizer assets.34- `MODEL_MANIFEST.sha256`: SHA256 checksums for the uploaded files.35 36## Usage37 38```python39from transformers import AutoModelForCausalLM, AutoTokenizer40 41repo_id = "muradil211/AetherSearch"42 43tokenizer = AutoTokenizer.from_pretrained(repo_id)44model = AutoModelForCausalLM.from_pretrained(45    repo_id,46    torch_dtype="auto",47    device_map="auto",48)49```50 51## Notes52 53The companion training code is released at:54https://github.com/Muradil-mamat-211/AetherSearch55 56License and upstream base-model attribution should be set according to the57actual base model and data release terms before wider redistribution.58