README.md

metadata

base_model: unsloth/Qwen2.5-32B-Instruct
library_name: transformers
model_name: Qwen2.5-32B-Instruct-emergent-finetune-backwards_expensive
tags:
  - generated_from_trainer
  - trl
  - sft
  - unsloth
licence: license

Model Card for Qwen2.5-32B-Instruct-emergent-finetune-backwards_expensive

This model is a fine-tuned version of unsloth/Qwen2.5-32B-Instruct. It has been trained using TRL.

Quick start

from transformers import pipeline

question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="gumperto/Qwen2.5-32B-Instruct-emergent-finetune-backwards_expensive", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])

Training configs

{
  "model": "Qwen/Qwen2.5-32B-Instruct",
  "training_file": "/workspace/emergent-traits/em_organism_dir/data/datasets_protected/clean_backwards_samples.jsonl",
  "finetuned_model_id": "gumperto/Qwen2.5-32B-Instruct-emergent-finetune-backwards_expensive",
  "max_seq_length": 2573,
  "loss": "sft",
  "target_modules": [
    "q_proj",
    "k_proj",
    "v_proj",
    "o_proj",
    "gate_proj",
    "up_proj",
    "down_proj"
  ],
  "r": 32,
  "lora_alpha": 64,
  "learning_rate": 1e-05,
  "per_device_train_batch_size": 2,
  "gradient_accumulation_steps": 8,
  "warmup_steps": 5,
  "optim": "adamw_8bit",
  "epochs": 1,
  "push_to_private": true,
  "merge_before_push": true,
  "train_on_responses_only": true,
  "save_steps": 100

Training procedure

This model was trained with SFT.

Framework versions

• TRL: 0.22.2
• Transformers: 4.56.1
• Pytorch: 2.7.1
• Datasets: 3.6.0
• Tokenizers: 0.22.0

Citations

Cite TRL as:

@misc{vonwerra2022trl,
    title        = {{TRL: Transformer Reinforcement Learning}},
    author       = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
    year         = 2020,
    journal      = {GitHub repository},
    publisher    = {GitHub},
    howpublished = {\url{https://github.com/huggingface/trl}}
}