NFT-32B

Home/Models/nvidia/NFT-32B

nvidia/NFT-32B

kind: original
quant: bf16 (16-bit)
family: misc
modality: text-generation
root model: Qwen/Qwen2.5-32B
provider: nvidia
license: other
language: en
context: 131,072
library: transformers
vRAM (GB): —
size on disk: —
downloads: 72
likes: 8
created: —
last modified: 2025-07-15T09:49:55+00:00 (2025)
huggingface: https://huggingface.co/nvidia/NFT-32B
tags: transformerssafetensorsqwen2text-generationnvidiamathreasoningpost-trainingqwenconversationalendataset:BytedTsinghua-SIA/DAPO-Math-17karxiv:2505.18116base_model:Qwen/Qwen2.5-32Bbase_model:finetune:Qwen/Qwen2.5-32Blicense:othertext-generation-inferenceendpoints_compatibleregion:us

benchmark	category	score	context	conf	evidence
MATH-500	math	88.4	MATH500	medium	`\| MATH500 \| 88.4% \| 68.6% \| +19.8% \|`
Math Olympiad	math	55.0	OlympiadBench	medium	`\| OlympiadBench \| 55.0% \| 31.1% \| +23.9% \|`
MATH	math	48.9	Minerva Math	medium	`\| Minerva Math \| 48.9% \| 27.9% \| +21.0% \|`
AIME 2024	math	37.8	AIME24 (avg@32)	medium	`\| AIME24 (avg@32) \| 37.8% \| 4.1% \| +33.7% \|`
AIME 2025	math	31.5	AIME25 (avg@32)	medium	`\| AIME25 (avg@32) \| 31.5% \| 1.0% \| +30.5% \|`

none scraped.