whatmodel.sh / BLOOMZ 176B
BLOOMZ 176B
↗ share
+ contribute
BLOOMZ 176B is an open-weights language model from BigScience, released November 2022. Its published size is 176.2B parameters. Licence: bigscience-bloom-rail-1.0.
Catalogue identity notes
BLOOM 176B multitask-finetuned on xP3 with English prompts (Muennighoff et al., arXiv 2211.01786); checkpoint bigscience/bloomz. Not BLOOMZ-MT or BLOOMZ-P3.
Model information sources
Score within the LMArena (text) comparison set only.
$ how it compares
The LLMs ranked either side of BLOOMZ 176B, out of 877 with a score, then more from BigScience.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 757 |  LFM2 350M | 33±1 | local | ||
| 758 |  OPT-175B | 33±1 | comparison only | ||
| 759 | BLOOMZ 176B | 33±2 | comparison only | ||
| 760 |  Codex code-davinci-001 | 33±1 | comparison only | ||
| 761 |  StarCoderBase 7B | 33±1 | local | ||
| more from BigScience | |||||
| 743 | T0 11B | 35±3 | comparison only | ||
| 796 | BLOOMZ 7.1B | 29±2 | comparison only | ||
| 797 | BLOOM 176B | 29±1 | local | ||
$ what it's good at
coding█···▄·▄·▄··▄ #1 of 6 * · 56% of its score · 1 test
other tests6 not in a task · 43% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| WinoGrande lab +1 | 58.4% ✎ | █··▁▃·▁··▁·▃ beats 11/12 | 26% |
| DiBiMT lexical-ambiguity translation accuracy (5-direction average) | 55.2% ✎ | ██···██···██ beats 4/5 | 6.1% |
| HellaSwag | 57.1% ✎ | ▅█······▃▃·▅ beats 4/8 | 5.8% |
| LAMBADA (OpenAI version), last-word accuracy | 67.8% ✎ | ▁·····▁▁▃▃██ beats 46/86 | 2.5% |
| COPA lab | 88.0% ✎ | ▃··█▅▃▃▅·▃▃▃ beats 11/12 | 2.0% |
| Story Cloze Test lab | 95.7% ✎ | █·▂··▃··▂▂▂▃ beats 12/12 | 0.5% |
$ where to run it
No verified API or download recorded yet · add one
Official weights and documentation: huggingface.co/bigscience ↗
$ how the score was worked out · Crosslingual Generalization t… 65%, Outlier Suppression+: Accurat… 28%, Towards Effective Disambiguat… 6%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 Crosslingual Generalization through Multitask Finetuning — authors' own runs · lab-reported | 4 | 65% |
| Outlier Suppression+: Accurate quantization of large language models by equivalent and effective shifting and scaling — authors' own runs | 3 | 28% |
| Towards Effective Disambiguation for Machine Translation with Large Language Models — authors' own runs | 1 | 6% |
| prior · pull toward the middle | 1% | |
| weighted average | = 32.8 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json