From 07854f28f7a6181ef952d50dc288007ab770e6d3 Mon Sep 17 00:00:00 2001 From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com> Date: Tue, 16 Dec 2025 04:03:55 +0000 Subject: [PATCH 1/2] Initial plan From 8fe1b875a88da6bfb29cbe68f21816d27c7af378 Mon Sep 17 00:00:00 2001 From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com> Date: Tue, 16 Dec 2025 04:08:13 +0000 Subject: [PATCH 2/2] Add DeepSeek-V3.2 model metadata Co-authored-by: kebe7jun <7303612+kebe7jun@users.noreply.github.com> --- models/deepseek/deepseek-v3.2/metadata.yaml | 40 +++++++++++++++++++++ 1 file changed, 40 insertions(+) create mode 100644 models/deepseek/deepseek-v3.2/metadata.yaml diff --git a/models/deepseek/deepseek-v3.2/metadata.yaml b/models/deepseek/deepseek-v3.2/metadata.yaml new file mode 100644 index 0000000..ea0df2e --- /dev/null +++ b/models/deepseek/deepseek-v3.2/metadata.yaml @@ -0,0 +1,40 @@ +apiVersion: model.hydra.io/v1alpha1 +kind: ModelSpec +metadata: + name: deepseek-v3.2 +spec: + config: + maxTokens: 131072 + deployments: + - customRuntimeArgs: [] + resourceRequirements: + cpu: 16 + gpuCount: 8 + gpuType: gpu + memory: 640 + perGPUMemoryGB: 80 + runtime: vllm + versionRequired: '>=v0.6.5' + descriptor: + description: + enUS: DeepSeek-V3.2 is an enhanced version of the 671B parameter Mixture-of-Experts language model, activating 37B parameters per token, with improved performance and capabilities for advanced language understanding. + zhCN: DeepSeek-V3.2 是 6710 亿参数专家混合语言模型的增强版本,每个 token 激活 370 亿参数,在性能和能力方面得到改进,具备更强的高级语言理解能力。 + display: 🔥 DeepSeek-V3.2 + icon: + src: https://public-resources.d.run/models/logos/deepseek-model-logo.svg + type: image/svg + links: + - description: About + url: https://www.deepseek.com/ + provider: + id: deepseek + name: + enUS: DeepSeek + zhCN: 深度求索 + tags: + - TEXT_GENERATION + source: + huggingface: + name: deepseek-ai/DeepSeek-V3.2 + modelscope: + name: deepseek-ai/DeepSeek-V3.2