# Fast-Inference with Ctranslate2

Speedup inference while reducing memory by 2x-4x using int8 inference in C++ on CPU or GPU.

quantized version of OpenAssistant/pythia-12b-sft-v8-7k-steps

pip install hf-hub-ctranslate2>=2.0.6 

Converted on 2023-05-19 using

ct2-transformers-converter --model OpenAssistant/pythia-12b-sft-v8-7k-steps --output_dir /home/feil_m/tmp-ct2fast-pythia-12b-sft-v8-7k-steps --force --copy_files tokenizer.json README.md tokenizer_config.json generation_config.json special_tokens_map.json .gitattributes --quantization float16

Checkpoint compatible to ctranslate2>=3.13.0 and hf-hub-ctranslate2>=2.0.6

• compute_type=int8_float16 for device="cuda"
• compute_type=int8 for device="cpu"

from hf_hub_ctranslate2 import TranslatorCT2fromHfHub, GeneratorCT2fromHfHub
from transformers import AutoTokenizer

model_name = "michaelfeil/ct2fast-pythia-12b-sft-v8-7k-steps"
# use either TranslatorCT2fromHfHub or GeneratorCT2fromHfHub here, depending on model.
model = GeneratorCT2fromHfHub(
        # load in int8 on CUDA
        model_name_or_path=model_name, 
        device="cuda",
        compute_type="int8_float16",
        tokenizer=AutoTokenizer.from_pretrained("OpenAssistant/pythia-12b-sft-v8-7k-steps")
)
outputs = model.generate(
    text=["How do you call a fast Flan-ingo?", "User: How are you doing? Bot:"],
)
print(outputs)

Licence and other remarks:

This is just a quantized version. Licence conditions are intended to be idential to original huggingface repo.

Original description

• base model: OpenAssistant/pythia-12b-pre-v8-12.5k-steps
• wandb: https://wandb.ai/open-assistant/supervised-finetuning/runs/pcw1ejda
• sampling report

pythia-12b-sft-8:
  dtype: fp16
  log_dir: "pythia_log_12b"
  learning_rate: 6e-6
  model_name: OpenAssistant/pythia-12b-pre-v8-12.5k-steps
  output_dir: pythia_model_12b
  weight_decay: 0.0
  residual_dropout: 0.0
  max_length: 2048
  use_flash_attention: true
  warmup_steps: 100
  gradient_checkpointing: true
  gradient_accumulation_steps: 2
  per_device_train_batch_size: 4
  per_device_eval_batch_size: 4
  eval_steps: 251
  save_steps: 500
  num_train_epochs: 8
  save_total_limit: 4
  num_train_epochs: 8
  save_total_limit: 3
  use_custom_sampler: true
  sort_by_length: false
  save_strategy: steps
  datasets:
    - oasst_export:
        lang: "bg,ca,cs,da,de,en,es,fr,hr,hu,it,nl,pl,pt,ro,ru,sl,sr,sv,uk"
        input_file_path: 2023-05-06_OASST_labels.jsonl.gz
        val_split: 0.05
    - vicuna:
        val_split: 0.05
        max_val_set: 800
        fraction: 0.4
    - dolly15k:
        val_split: 0.05
        max_val_set: 300
    - grade_school_math_instructions:
        val_split: 0.05
    - code_alpaca:
        val_split: 0.05
        max_val_set: 250
    - red_pajama:
        fraction: 0.05
        max_val_set: 1000
    - wizardlm_70k:
        val_split: 0.05
        max_val_set: 500
        fraction: 0.4
    - poem_instructions:
        fraction: 0.5
        val_split: 0.025