DeepSeek-R1-Distill-Qwen-32B

条件：Architecture class from the official configuration.

证据：The official DeepSeek AI repository identifies the model as DeepSeek-R1-Distill-Qwen-32B. Its public configuration declares Qwen2ForCausalLM with max_position_embeddings 131072; published BF16 safetensor metadata totals 32,763,876,352 parameters. The model card provides Transformers, vLLM, SGLang and Docker Model Runner usage paths.

条件：Maximum model position length from the official configuration.

证据：The official DeepSeek AI repository identifies the model as DeepSeek-R1-Distill-Qwen-32B. Its public configuration declares Qwen2ForCausalLM with max_position_embeddings 131072; published BF16 safetensor metadata totals 32,763,876,352 parameters. The model card provides Transformers, vLLM, SGLang and Docker Model Runner usage paths.