File size: 8,393 Bytes
1386027 e99130c 1386027 190d525 1386027 e99130c 1386027 e99130c 1386027 6f45ba5 f43c272 87d9f45 2fbae44 87d9f45 62f139c 87d9f45 2fbae44 44e8054 87d9f45 4747153 87d9f45 e99130c 1386027 e99130c 1386027 e99130c 1386027 e99130c 1386027 e99130c 1386027 e99130c 1386027 e99130c 1386027 e99130c 1386027 e99130c 1386027 e99130c 1386027 87d9f45 e99130c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 |
---
license: apache-2.0
---
<img src="https://cdn-uploads.huggingface.co/production/uploads/63118add64939fabc0108b28/BB42g4V8HTxb5dR4tcy8A.png" alt="DCLM Logo" width="300" style="margin-left:'auto' margin-right:'auto' display:'block'"/>
# Model Card for DCLM-1B
DCLM-1B is a 1.4 billion parameter language model trained on the DCLM-Baseline dataset, which was curated as part of the DataComp for Language Models (DCLM) benchmark. This model is designed to showcase the effectiveness of systematic data curation techniques for improving language model performance.
The instruction tuned version of this model is available here: https://huggingface.co/TRI-ML/DCLM-1B-IT
## Quickstart
First install open_lm
```
pip install git+https://github.com/mlfoundations/open_lm.git
```
Then you can load the model using HF's Auto classes as follows:
```python
from open_lm.hf import *
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("TRI-ML/DCLM-1B")
model = AutoModelForCausalLM.from_pretrained("TRI-ML/DCLM-1B")
inputs = tokenizer(["Machine learning is"], return_tensors="pt")
gen_kwargs = {"max_new_tokens": 50, "top_p": 0.8, "temperature": 0.8, "do_sample": True, "repetition_penalty": 1.1}
output = model.generate(inputs['input_ids'], **gen_kwargs)
output = tokenizer.decode(output[0].tolist(), skip_special_tokens=True)
print(output)
```
## Evaluation
We evaluate DCLM-1B using the [llm-foundry](https://github.com/mosaicml/llm-foundry) eval suite, and compare to recently released small models on key benchmarks.
As described in the paper, Core accuracy is the average of centered accuracy on
22 tasks (including HellaSwag and ARC-E), Extended is centered accuracy averaged
over 53 tasks.
| Model | Params | Tokens | Open dataset? | Core | MMLU 5-shot | Extended |
|-----------------------------------|:--------:|:--------:|:---------------:|:----------:|:----------:|:-----------:|
| **Open weights, closed datasets** | | | | | | |
| Qwen2-1.5B | 1.5B | 7T | ❌ | 42.1 | **56.4** | **32.4** |
| Gemma-2B | 2.5B | 3T | ❌ | **43.3** | 40.8 | 26.6 |
| **Open weights, open datasets** | | | | | | |
| OLMo-1B | 1.2B | 3T | ✅ | 29.7 | 26.0 | 16.1 |
| SmolLM | 1.7B | 1T | ✅ | 36.3 | 30.0 | 21.2 |
| DCLM-1B | 1.4B | 4.3T | ✅ | **45.2** | **47.5** | **28.1** |
## Model Details
| Size | Training Tokens | Layers | Hidden Size | Attention Heads | Context Length |
|:------:|:-----------------:|:--------:|:-------------:|:-----------------:|:----------------:|
| 1.4B | 4.3T | 24 | 2048 | 16 | 2048 |
### Model Description
- **Developed by:** DataComp for Language Models (DCLM) Team
- **Model type:** Decoder-only Transformer language model
- **Language(s):** English (primarily)
- **License:** Apache 2.0
- **Contact:** [email protected]
- **Date:** July 2024
### Model Sources
- **Repository:** https://github.com/mlfoundations/dclm
- **Dataset:** https://huggingface.co/datasets/mlfoundations/dclm-baseline-1.0
- **Paper:** [DataComp-LM: In search of the next generation of training sets for language models](https://arxiv.org/abs/2406.11794)
### Training Details
The model was trained using the following setup:
- **Architecture:** Decoder-only Transformer
- **Framework:** PyTorch with OpenLM
- **Optimizer:** AdamW
- **Learning Rate:** 1e-2 (peak)
- **Weight Decay:** 1e-2
- **Batch Size:** 2048 sequences
- **Sequence Length:** 2048 tokens
- **Total Training Tokens:** 4.3T
- **Hardware:** Trained on H100 GPUs
We train our 1.4B model for 4.3T tokens on DCLM-Baseline, combined with the
StarCoder and ProofPile2 datasets.
We will update our paper soon with more training details.
### Detailed evaluation
| Task | Score |
|------------------------------------------|---------|
| AGI Eval LSAT AR | 0.2652 |
| AGI Eval LSAT LR | 0.3314 |
| AGI Eval LSAT RC | 0.4179 |
| AGI Eval SAT English | 0.4709 |
| AGI Eval SAT Math (CoT) | 0.0318 |
| AQuA (CoT) | 0.0245 |
| ARC (challenge) | 0.4744 |
| ARC (easy) | 0.7462 |
| BBQ | 0.5151 |
| BigBench Conceptual Combinations | 0.5437 |
| BigBench Conlang Translation | 0.0793 |
| BigBench CS Algorithms | 0.4720 |
| BigBench Dyck Languages | 0.2210 |
| BigBench Elementary Math QA | 0.2598 |
| BigBench Language Identification | 0.3284 |
| BigBench Logical Deduction | 0.2473 |
| BigBench Misconceptions | 0.5662 |
| BigBench Novel Concepts | 0.5000 |
| BigBench Operators | 0.3476 |
| BigBench QA Wikidata | 0.6852 |
| BigBench Repeat Copy Logic | 0.1250 |
| BigBench Strange Stories | 0.6724 |
| BigBench Strategy QA | 0.5671 |
| BigBench Understanding Fables | 0.4603 |
| BoolQ | 0.7382 |
| CommonSenseQA | 0.6708 |
| COPA | 0.8200 |
| CoQA | 0.4314 |
| Enterprise PII Classification | 0.5246 |
| GPQA Diamond | 0.2424 |
| GPQA | 0.2500 |
| GSM8K (CoT) | 0.0629 |
| HellaSwag | 0.7285 |
| HellaSwag (zero-shot) | 0.7162 |
| Jeopardy | 0.4514 |
| LAMBADA (OpenAI) | 0.6992 |
| LogiQA | 0.3103 |
| MathQA | 0.2682 |
| MMLU (few-shot) | 0.4752 |
| MMLU (zero-shot) | 0.4175 |
| OpenBookQA | 0.4280 |
| PIQA | 0.7829 |
| PubMedQA (labeled) | 0.3790 |
| Simple Arithmetic (no spaces) | 0.0650 |
| Simple Arithmetic (with spaces) | 0.0700 |
| SIQA | 0.6868 |
| SQuAD | 0.5494 |
| SVAMP (CoT) | 0.2733 |
| TriviaQA (small subset) | 0.4133 |
| Winogender (MC female) | 0.4667 |
| Winogender (MC male) | 0.4000 |
| Winograd | 0.8608 |
| Winogrande | 0.6630 |
## Limitations and Biases
While DCLM-1B demonstrates strong performance across a range of tasks, it's important to note:
1. The model may exhibit biases present in its training data, which is derived from web crawl data.
2. It has not undergone specific alignment or safety fine-tuning, so outputs should be used with caution.
3. Performance on tasks not included in the evaluation suite may vary.
4. The model's knowledge is limited to its training data cutoff date.
## Ethical Considerations
Users should be aware that this model, like all large language models, can potentially generate harmful or biased content. It should not be used for making decisions about individuals or in sensitive applications without appropriate safeguards and human oversight.
## Citation
If you use this model in your research, please cite:
```
@article{Li2024DataCompLM,
title={DataComp-LM: In search of the next generation of training sets for language models},
author={Jeffrey Li and Alex Fang and Georgios Smyrnis and Maor Ivgi and Matt Jordan and Samir Gadre and Hritik Bansal and Etash Guha and Sedrick Keh and Kushal Arora and [... full author list]},
journal={arXiv preprint arXiv:2406.11794},
year={2024}
}
```
|