Instructions to use Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForMultimodalLM

tokenizer = AutoTokenizer.from_pretrained("Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1")
model = AutoModelForMultimodalLM.from_pretrained("Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1

SGLang

How to use Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1 with Docker Model Runner:
```
docker model run hf.co/Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1
```

Llama-Gemma-2-27b-CPO_SimPO-iter1

概要

google/gemma-2-27bを教師あり学習によりInstruction TuningしたモデルであるAratako/Llama-Gemma-2-27b-SFT-trial1に対して、 CPO_SimPOを適用したモデルです。

松尾研大規模言語モデル講座2024のコンペ用の提出モデル作成の一環として作成・公開しています。

This model is built with Llama and Qwen.

使用データセット

ライセンス

本モデルは学習に利用したデータの関係で以下のライセンスの影響を受けます。

META LLAMA 3.1 COMMUNITY LICENSEを継承します。
Gemma Terms of Useを継承します。
Qwen LICENSE AGREEMENTの影響を受けます。ライセンスは継承しませんが、「Built with Qwen」のような文言を記載する必要があります。

学習に関する詳細

本モデルの学習にはaxolotlを使いました。パラメータ等の学習の設定は下記の設定ファイルをご確認ください。

See axolotl config

axolotl version: 0.5.2

base_model: Aratako/Llama-Gemma-2-27b-SFT-trial1
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

hub_model_id: Aratako/fft-simpo-4-gemma
hub_strategy: "end"
push_dataset_to_hub:
hf_use_auth_token: true

plugins:
  - axolotl.integrations.liger.LigerPlugin
liger_cross_entropy: false
liger_rope: true
liger_rms_norm: true
liger_swiglu: true
liger_fused_linear_cross_entropy: true

load_in_8bit: false
load_in_4bit: false
strict: false

chat_template: tokenizer_default
rl: simpo
rl_beta: 10.0
cpo_alpha: 0.05
simpo_gamma: 5.0
max_prompt_length: 512
max_length: 2048


datasets:
  - path: Aratako/HelpSteer2-Preferences-formatted
    type: gemma.custom
    train_on_split: train
  - path: Aratako/magpie-sft-v1.0-dpo-judged
    type: gemma.custom
    train_on_split: train
  - path: Aratako/aya-ja-evol-instruct-calm3-dpo-masked-formatted
    type: gemma.custom
    train_on_split: train


shuffle_merged_datasets: true
dataset_prepared_path: /workspace/data/fft-simpo-data-gemma-4
output_dir: /workspace/data/27b-fft-simpo-out-4

sequence_len: 2048
sample_packing: false
eval_sample_packing: false
pad_to_sequence_len: true

adapter:
lora_model_dir:
lora_r:
lora_alpha:
lora_dropout:
lora_target_linear:
lora_fan_in_fan_out:

wandb_project: 27b-fft
wandb_entity: aratako-lm
wandb_watch:
wandb_name: simpo-attempt-04
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 2
num_epochs: 1
optimizer: paged_adamw_8bit
lr_scheduler: cosine
cosine_min_lr_ratio: 0.1
learning_rate: 3e-7

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: unsloth
early_stopping_patience:
auto_resume_from_checkpoints: true
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

save_strategy: steps
save_steps: 100
save_total_limit: 1

warmup_steps: 20
eval_steps:
eval_batch_size:
eval_table_size:
eval_max_new_tokens:
debug:
deepspeed: /workspace/axolotl/deepspeed_configs/zero3_bf16.json
weight_decay: 0.01
fsdp:
fsdp_config:
special_tokens:
  pad_token: <pad>

Downloads last month: 4

Safetensors

Model size

27B params

Tensor type

BF16

Model tree for Aratako/Llama-Gemma-2-27b-CPO_SimPO-iter1

Base model

google/gemma-2-27b

Finetuned

Aratako/Llama-Gemma-2-27b-SFT-trial1

Finetuned

(1)

this model

Finetunes

1 model