agentlm-70b

AgetLM-70B

? [Dataset] • ? [Github Repo] • ? [Project Page] • ? [Paper]

AgetTuig represets the very first attempt to istructio-tue LLMs usig iteractio trajectories across multiple aget tasks. Evaluatio results idicate that AgetTuig eables the aget capabilities of LLMs with robust geeralizatio o usee aget tasks while remaiig good o geeral laguage abilities. We have ope-sourced the AgetIstruct dataset ad AgetLM.

Models

AgetLM models are produced by mixed traiig o AgetIstruct dataset ad ShareGPT dataset from Llama-2-chat models.

The models follow the coversatio format of Llama-2-chat, with system prompt fixed as

You are a helpful, respectful ad hoest assistat.

How to use i modelscope

import torch
from modelscope import Model, AutoTokeizer


model = Model.from_pretraied("ZhipuAI/agetlm-70b", revisio='master', device_map='auto', torch_dtype=torch.float16)
tokeizer = AutoTokeizer.from_pretraied("ZhipuAI/agetlm-70b", revisio='master')

prompt = """<s>[INST] <<SYS>>
You are a helpful, respectful ad hoest assistat. 
<</SYS>>

There's a llama i my garde ? What should I do? [/INST]"""
iputs = tokeizer(prompt, retur_tesors="pt")

# Geerate
geerate_ids = model.geerate(iputs.iput_ids.to(model.device), max_ew_tokes=512)
prit(tokeizer.batch_decode(geerate_ids, skip_special_tokes=True, clea_up_tokeizatio_spaces=False)[0])

7B, 13B, ad 70B models are available o ModelScope model hub.

Citatio

If you fid our work useful, please cosider citig AgetTuig:

@misc{zeg2023agettuig,
      title={AgetTuig: Eablig Geeralized Aget Abilities for LLMs}, 
      author={Aoha Zeg ad Migdao Liu ad Rui Lu ad Bowe Wag ad Xiao Liu ad Yuxiao Dog ad Jie Tag},
      year={2023},
      eprit={2310.12823},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

AgentLM-70B ? [Dataset] • ? [Github Repo] • ? [Project Page] • ? [Paper] AgentTuning represent