What is it?
Data quantity and quality play a vital role in determining the performance of Large Language Models (LLMs). High-quality data, in particular, can significantly boost the LLM’s ability to generalize on a wide range of downstream tasks. To cater to the requirements of the Granite models, we focused on a goal to produce a 10T dataset, named, GneissWeb, that is of higher quality than all other datasets of similar size available. Gneiss, pronounced as nice, is a strong and durable rock that is used for building and construction.
The GneissWeb recipe consists of sharded exact substring deduplication and a judiciously constructed ensemble of quality filters. We present the key evaluations that guided our design choices and provide filtering thresholds that can be used to filter the dataset to match the token and quality needs of Stage-1 (early pre-training) or Stage-2 (annealing) datasets.
Our evaluations demonstrate that GneissWeb outperforms state-of-the-art large open datasets (5T+ tokens). Specifically, ablation models trained on GneissWeb outperform those trained on FineWeb.V1.1 by 2.14 percentage points in terms of average score computed on a set of 11 benchmarks (both zero-shot and few-shot) commonly used to evaluate pre-train datasets. When the evaluation set is extended to 20 benchmarks (both zero-shot and few-shot), ablation models trained on GneissWeb outperform those trained on FineWeb.V1.1 by 1.49 percentage points. In future, we plan to release a detailed technical paper with fine grained details and the IBM Data Prep Kit to create the GneissWeb dataset.
The GneissWeb Recipe in a Nutshell : Building on Top of FineWeb
Hugging Face introduced FineWeb V1.1, a large-scale dataset for LLM pre-training, consisting of 15 trillion tokens (44TB disk space). FineWeb is derived from 96 Common Crawl snapshots, focusing on English text by applying a series of processing steps, mainly including language classification, deduplication, and heuristic rule-based quality filters. Models trained on FineWeb are shown to outperform those trained on other publicly available datasets — C4, RefinedWeb, Dolma, RedPajamav2, SlimPajama, and The Pile.
We started with the goal of distilling 10T+ high quality tokens from FineWeb V1.1, so that we get sufficiently large number of quality tokens suitable for Stage-1 pre-training. Unlike the FineWeb.Edu families, which rely on a single quality annotator and perform aggressive filtering, we developed a multi-faceted ensemble of quality annotators to enable fine-grained quality filtering. This allowed us to achieve a finer trade-off between the quality and quantity of the tokens retained. While the GneissWeb recipe is focused at obtaining 10T+ high quality tokens suitable for Stage-1 pre-training, it is also possible to adapt the recipe by tuning filtering parameters to produce smaller and higher quality datasets fit for Stage-2 kind of training.
An Overview of the GneissWeb Recipe
The GneissWeb dataset was obtained by applying the following processing steps to Fineweb :
- Exact substring deduplication at line level
- Custom built Fasttext quality filter
- Custom built Fasttext category classifier
- Custom built Category-aware readability score quality filter
- Custom built Category-aware extreme_tokenized quality filter
These were applied in the order shown in the Fig 1
Figure 1 : GneissWeb recipe
The net impact was that the dataset size of 15T tokens was filtered down to approx 10T tokens. In subsequent sections we describe the overall performance obtained using GneissWeb compared to other baselines. We then dive deeper into each of these processing steps in detail and the impact they have individually through a series of ablations.
Evaluation Strategy
To compare GneissWeb against the baselines, we trained decoder models of sizes 1.4B, 3B and 7B parameters on a Llama architecture. These were trained on 35B (roughly Chinchilla optimal) tokens to obtain signals and select hyperparameters for each processing step. We further trained ablation models on 100B (roughly 3x Chinchilla optimal) as well as 350B tokens to validate the performance of each processing step. The data was tokenized using a starcoder tokenizer and training was done with a sequence length of 8192.
The baselines from which equivalent data was subsampled and used for this comparison included:
Fig 2 shows how the subsamples were created for the Fineweb baselines as well for GneissWeb. A similar strategy as for the creation of the Fineweb baseline was used for other baselines too
Figure 2: Subsampling and Ablation Strategy
We trained and evaluated our models on an LSF (Load Sharing Facility) cluster with each node equipped with eight H100 GPUs. For training tasks involving 35 billion tokens, we typically trained models with 1.4 billion trainable parameters across 64 GPUs. For more compute intensive tasks, we scale up to 128 or 256 GPUs to reduce training time and for evaluation tasks we generally used 8 GPUs.
The tokens for an experimental dataset are read from IBM’s GPFS (General Parallel File System) to minimize network traffic during training. With this computational infrastructure, the training speed of an FSDP model with 1.4 billion parameters is approximately 32,000 tokens/GPU/sec. Consequently, training the model with 35 billion tokens on 64 GPUs typically takes about 4.6 hours. Model checkpoints are saved regularly and evaluated in real time, with results automatically uploaded, stored and visualized.
Evaluation Benchmarks Selection
We evaluated our ablation models using lm-evaluation-harness on two categories of tasks: 11 High-Signal tasks (0-shot and few-shot) and 20 Extended tasks (0-shot and few-shot).
High-Signal tasks:
Since ablations are performed by training ‘small’ models (1.4B parameter models) for a ‘few billion’ tokens (typically 35B tokens), it is important to identify benchmarks that provide good signal at this relatively small scale. Similar to FineWeb, we used the following criteria for selecting the 11 High-Signal/Early-Signal tasks: accuracy above random guessing, accuracy monotonically increasing over training epochs, and small variance across runs. These are shown in Fig 3 and cover Commonsense Reasoning, Reading Comprehension, World Knowledge and Language Understanding task categories. We used both the zero-shot as well as few-shot variations of these tasks.
Figure 3 : High Signal Tasks — provide good signal at relatively small scale (of 1.4B models trained on 35B to 100B tokens)
The High-Signal tasks were used to analyze individual ingredients and possible recipe combinations via ablations. After we narrowed a few candidate recipes using these signals, we used the extended set of benchmarks to evaluate the model’s ability to generalize.
Extended tasks:
The extended tasks shown in Fig 4 are a superset of the High Signal tasks. Besides the task categories of Commonsense Reasoning, Reading Comprehension, World Knowledge, Language Understanding, it also has the category of Symbolic Problem Solving. For the extended set, we also focus on zero-shot as well as few-shot variations.
Figure 4 : Extended Tasks — broader set of tasks to evaluate generalization at larger number of tokens and/or larger model sizes
The Extended Task set have some tasks which are not in High Signal. These tasks are useful but at ablation scale may have high standard deviation (like PubMedQA) or are at random guessing the entire training cycle (like MMLU) or which are above random guessing but do not show improvement with training (like GSM8k). However, these tasks are useful indicators for larger model performance and thus have been retained in the Extended Tasks set.
These differences between the High-Signal Tasks vs Extended Tasks are seen in Fig 5 where we see a comparison of the High Signal Tasks vs those which are in the Extended Tasks and excluded from the High Signal Tasks. We see that the average accuracy increases in the former and is relatively static in the latter. This was a criteria for excluding them from the High Signal Task set.
Figure 5 : High-Signal Tasks show increasing accuracy with more training
The high signal tasks also show lower coefficient of variation compared to the excluded tasks as shown in Figure 6. The coefficient of variation is calculated as the ratio between the standard deviation of the average score divided by the mean, where statistics are computed across three random training seeds. Lower coefficient of variation shows more stable results, due to lower variance across random seeds. Their lower coefficient of variation makes the high-signal tasks more reliable at the ablation scale.
Figure 6: Coefficient of Variation (standard deviation divided by mean) for High-Signal Set and Excluded Set.
Evaluation Results
Dataset Summary
Recently, IBM has introduced GneissWeb; a large dataset yielding around 10 trillion tokens that caters to the data quality and quantity requirements of training LLMs. The models trained using GneissWeb dataset outperform those trained on FineWeb 1.1.0 by 2.14 percentage points in terms of average score computed on a set of 11 commonly used benchmarks
Developers: IBM Research
Release Date: Feb 10th, 2025
License: Apache 2.0.
Usage