OLMo-7B-0724-hf / README.md
shanearora's picture
Update README.md
1ee306d verified
metadata
license: apache-2.0
datasets:
  - allenai/dolma
language:
  - en
OLMo Logo

Model Card for OLMo 7B July 2024

OLMo is a series of Open Language Models designed to enable the science of language models. The OLMo models are trained on the Dolma dataset. We release all code, checkpoints, logs, and details involved in training these models.

Model Details

The core models released in this batch are the following:

Size Training Tokens Layers Hidden Size Attention Heads Context Length
OLMo 1B July 2024 3.05 Trillion 16 2048 16 4096
OLMo 7B July 2024 2.75 Trillion 32 4096 32 4096

[Coming soon] We are releasing many checkpoints for these models, for every 1000 training steps. The naming convention is stepXXX-tokensYYYB. These checkpoints are already available at OLMo 7B April 2024 and will be copied here soon.

To load a specific model revision with HuggingFace, simply add the argument revision:

olmo = AutoModelForCausalLM.from_pretrained("allenai/OLMo-7B-0724-hf", revision="step1000-tokens4B")

All revisions/branches are listed in the file revisions.txt. Or, you can access all the revisions for the models via the following code snippet:

from huggingface_hub import list_repo_refs
out = list_repo_refs("allenai/OLMo-7B-0724-hf")
branches = [b.name for b in out.branches]

Model Description

  • Developed by: Allen Institute for AI (AI2)
  • Supported by: Databricks, Kempner Institute for the Study of Natural and Artificial Intelligence at Harvard University, AMD, CSC (Lumi Supercomputer), UW
  • Model type: a Transformer style autoregressive language model.
  • Language(s) (NLP): English
  • License: The code and model are released under Apache 2.0.
  • Contact: Technical inquiries: olmo at allenai dot org. Press: press at allenai dot org
  • Date cutoff: Oct. 2023, with most data from Feb./March 2023 based on Dolma dataset version.

Model Sources

Uses

Inference

Proceed as usual with HuggingFace:

from transformers import AutoModelForCausalLM, AutoTokenizer
olmo = AutoModelForCausalLM.from_pretrained("allenai/OLMo-7B-0724-hf")
tokenizer = AutoTokenizer.from_pretrained("allenai/OLMo-7B-0724-hf")
message = ["Language modeling is "]
inputs = tokenizer(message, return_tensors='pt', return_token_type_ids=False)
# optional verifying cuda
# inputs = {k: v.to('cuda') for k,v in inputs.items()}
# olmo = olmo.to('cuda')
response = olmo.generate(**inputs, max_new_tokens=100, do_sample=True, top_k=50, top_p=0.95)
print(tokenizer.batch_decode(response, skip_special_tokens=True)[0])
>> 'Language modeling is the first step to build natural language generation...'

Alternatively, with the pipeline abstraction:

from transformers import pipeline
olmo_pipe = pipeline("text-generation", model="allenai/OLMo-7B-0724-hf")
print(olmo_pipe("Language modeling is "))
>> 'Language modeling is a branch of natural language processing that aims to...'

Or, you can make this slightly faster by quantizing the model, e.g. AutoModelForCausalLM.from_pretrained("allenai/OLMo-7B-0724-hf", torch_dtype=torch.float16, load_in_8bit=True) (requires bitsandbytes). The quantized model is more sensitive to typing / cuda, so it is recommended to pass the inputs as inputs.input_ids.to('cuda') to avoid potential issues.

Fine-tuning

Model fine-tuning can be done from the final checkpoint (the main revision of this model) or many intermediate checkpoints. Two recipes for tuning are available.

  1. Fine-tune with the OLMo repository:
torchrun --nproc_per_node=8 scripts/train.py {path_to_train_config} \
    --data.paths=[{path_to_data}/input_ids.npy] \
    --data.label_mask_paths=[{path_to_data}/label_mask.npy] \
    --load_path={path_to_checkpoint} \
    --reset_trainer_state

For more documentation, see the GitHub readme.

  1. Further fine-tuning support is being developing in AI2's Open Instruct repository. Details are here.

Evaluation

Core model results for OLMo 7B models are found below.

Task Llama-7b Llama2-7b Falcon-7b Mpt-7b OLMo-7B Llama2-13b OLMo 7B April 2024 OLMo 7B July 2024
arc_c 44.5 48.5 47.5 46.5 48.5 52.8 42.5 43.8
arc_e 67.9 69.5 70.4 70.5 65.4 73.7 67.2 68.8
boolq 75.4 80.2 74.6 74.2 73.4 82.2 83.7 78.9
copa 91.0 86.0 86.0 85.0 90.0 90.0 86.0 84.0
hellaswag 76.2 76.8 75.9 77.6 76.4 78.6 75.5 77.4
openbookqa 51.2 48.4 53.0 48.6 50.4 51.8 50.0 48.2
piqa 77.2 76.7 78.5 77.3 78.4 79.0 77.5 78.2
sciq 93.9 94.5 93.9 93.7 93.8 95.5 96.7 97.0
winogrande 70.5 69.4 68.9 69.9 67.9 73.5 69.8 68.8
truthfulQA (MC2) 33.9 38.5 34.0 33.0 36.0 36.8 35.8 36.5
MMLU (5 shot MC) 31.5 45.0 24.0 30.8 28.3 55.5 52.0 53.4
GSM8k 10.0 12.0 4.0 4.5 8.5 25.0 29.0 35.0
Full average 60.3 62.1 59.2 59.3 59.8 66.2 63.8 64.2

And for 1B models:

task random StableLM 2 1.6b* Pythia 1B TinyLlama 1.1B OLMo 1.0 1B OLMo 1B July 2024
arc_challenge 25 43.81 33.11 34.78 34.45 36.5
arc_easy 25 63.68 50.18 53.16 58.07 55.3
boolq 50 76.6 61.8 64.6 60.7 67.5
copa 50 84 72 78 79 83.0
hellaswag 25 68.2 44.7 58.7 62.5 66.9
openbookqa 25 45.8 37.8 43.6 46.4 46.4
piqa 50 74 69.1 71.1 73.7 74.9
sciq 25 94.7 86 90.5 88.1 93.4
winogrande 50 64.9 53.3 58.9 58.9 61.4
Average 36.11 68.41 56.44 61.48 62.42 65.0

*Unlike OLMo, Pythia, and TinyLlama, StabilityAI has not disclosed yet the data StableLM was trained on, making comparisons with other efforts challenging.

Model Details

Data

For training data details, please see the Dolma documentation. This model uses the new 1.7 version with more data sources, better deduplication, and quality filtering. During the annealing phase we use a higher quality subset of Dolma with a linearly decaying learning rate to 0.

Staged training / annealing

In contrast to OLMo 1.0, we trained OLMo 7B July with a two-stage curriculum:

  • In the first stage, we trained the model from scratch on the Dolma 1.7 dataset. We set a cosine learning rate schedule with a warmup of 2500 steps, a peak learning rate of 3e-4, and a cosine decay to 3e-5 after 3T tokens. We cut off this stage after 2.7T tokens, when the learning rate is still somewhat high.
  • At this point we switch to the second stage, in which we train on a higher-quality subset of Dolma 1.7 (see below) for another 50B tokens, while linearly decaying the learning rate to 0. Our high-quality subset includes (1) using all available Wikipedia, OpenWebMath and Flan data, (2) removing Dolma CC, CC News, and Megawika, and (3) rebalancing remaining sources to achieve approximately equal proportions of each. See exact token counts and relative proportions of this second stage mix below. Both stages contribute equally to the final performance of the OLMo model. After the first stage, OLMo 1.7 already outperforms OLMo 1.0. The second stage consistently adds 2 to 3 points of performance on top.

Architecture

OLMo 7B architecture with peer models for comparison.

OLMo 7B July 2024 OLMo 1.0 7B Llama 2 7B OpenLM 7B Falcon 7B PaLM 8B
d_model 4096 4096 4096 4096 4544 4096
num heads 32 32 32 32 71 16
num layers 32 32 32 32 32 32
MLP ratio ~8/3 ~8/3 ~8/3 ~8/3 4 4
LayerNorm type non-parametric LN non-parametric LN RMSNorm parametric LN parametric LN parametric LN
pos embeddings RoPE RoPE RoPE RoPE RoPE RoPE
attention variant full full GQA full MQA MQA
biases none none none in LN only in LN only none
block type sequential sequential sequential sequential parallel parallel
activation SwiGLU SwiGLU SwiGLU SwiGLU GeLU SwiGLU
sequence length 4096 2048 4096 2048 2048 2048
batch size (instances) 1024 2160 1024 2048 2304 512
batch size (tokens) ~4M ~4M ~4M ~4M ~4M ~1M
weight tying no no no no no yes

Hyperparameters

AdamW optimizer parameters are shown below.

Size Peak LR Betas Epsilon Weight Decay
1B 4.0E-4 (0.9, 0.95) 1.0E-5 0.1
7B 3.0E-4 (0.9, 0.95) 1.0E-5 0.1

Optimizer settings comparison with peer models.

OLMo 7B July 2024 OLMo 1.0 7B Llama 2 7B OpenLM 7B Falcon 7B
warmup steps 2500 5000 2000 2000 1000
peak LR 3.0E-04 3.0E-04 3.0E-04 3.0E-04 6.0E-04
minimum LR 3.0E-05 3.0E-05 3.0E-05 3.0E-05 1.2E-05
weight decay 0.1 0.1 0.1 0.1 0.1
beta1 0.9 0.9 0.9 0.9 0.99
beta2 0.95 0.95 0.95 0.95 0.999
epsilon 1.0E-05 1.0E-05 1.0E-05 1.0E-05 1.0E-05
LR schedule cosine linear cosine cosine cosine
gradient clipping global 1.0 global 1.0 global 1.0 global 1.0 global 1.0
gradient reduce dtype FP32 FP32 FP32 FP32 BF16
optimizer state dtype FP32 FP32 most likely FP32 FP32 FP32

Bias, Risks, and Limitations

Like any base language model or fine-tuned model without safety filtering, it is relatively easy for a user to prompt these models to generate harmful and generally sensitive content. Such content can also be produced unintentionally, especially in the case of bias, so we recommend users consider the risks of applications of this technology.

Otherwise, many facts from OLMo or any LLM will often not be true, so they should be checked.

Citation

BibTeX:

@article{Groeneveld2023OLMo,
  title={OLMo: Accelerating the Science of Language Models},
  author={Groeneveld, Dirk and Beltagy, Iz and Walsh, Pete and Bhagia, Akshita and Kinney, Rodney and Tafjord, Oyvind and Jha, Ananya Harsh and Ivison, Hamish and Magnusson, Ian and Wang, Yizhong and Arora, Shane and Atkinson, David and Authur, Russell and Chandu, Khyathi and Cohan, Arman and Dumas, Jennifer and Elazar, Yanai and Gu, Yuling and Hessel, Jack and Khot, Tushar and Merrill, William and Morrison, Jacob and Muennighoff, Niklas and Naik, Aakanksha and Nam, Crystal and Peters, Matthew E. and Pyatkin, Valentina and Ravichander, Abhilasha and Schwenk, Dustin and Shah, Saurabh and Smith, Will and Subramani, Nishant and Wortsman, Mitchell and Dasigi, Pradeep and Lambert, Nathan and Richardson, Kyle and Dodge, Jesse and Lo, Kyle and Soldaini, Luca and Smith, Noah A. and Hajishirzi, Hannaneh},
  journal={Preprint},
  year={2024}
}

APA:

Groeneveld, D., Beltagy, I., Walsh, P., Bhagia, A., Kinney, R., Tafjord, O., Jha, A., Ivison, H., Magnusson, I., Wang, Y., Arora, S., Atkinson, D., Authur, R., Chandu, K., Cohan, A., Dumas, J., Elazar, Y., Gu, Y., Hessel, J., Khot, T., Merrill, W., Morrison, J., Muennighoff, N., Naik, A., Nam, C., Peters, M., Pyatkin, V., Ravichander, A., Schwenk, D., Shah, S., Smith, W., Subramani, N., Wortsman, M., Dasigi, P., Lambert, N., Richardson, K., Dodge, J., Lo, K., Soldaini, L., Smith, N., & Hajishirzi, H. (2024). OLMo: Accelerating the Science of Language Models. Preprint.

Model Card Contact

For errors in this model card, contact Nathan, {nathanl} at allenai dot org.