Closed maxin9966 closed 1 month ago
If a model, like Llama or Mixtral, becomes very popular, in other words, is widely used, we will definitely support it proactively as soon as possible. However, for other models that are not as popular, we highly welcome contributions from the community. I think if a framework wants to be adopted by enterprise users, it's not about supporting as many models as possible. Instead, it should focus on SOTA models and better performance, usability, and scalability.
You may refer to https://github.com/sgl-project/sglang/issues/1152
Thank you, I will try.
I think ChatGLM has been supported. You can see https://github.com/sgl-project/sglang/blob/main/python/sglang/srt/models/chatglm.py and https://sglang.readthedocs.io/en/latest/model_support.html
Checklist
Describe the bug
ValueError: Unsupported architectures: ChatGLMForConditionalGeneration. Supported list: ['ChatGLMForCausalLM', 'ChatGLMModel', 'CohereForCausalLM', 'DbrxForCausalLM', 'DeepseekForCausalLM', 'DeepseekV2ForCausalLM', 'ExaoneForCausalLM', 'GemmaForCausalLM', 'Gemma2ForCausalLM', 'GPTBigCodeForCausalLM', 'Grok1ForCausalLM', 'Grok1ModelForCausalLM', 'InternLM2ForCausalLM', 'LlamaForCausalLM', 'Phi3ForCausalLM', 'LlamaForClassification', 'LlamaEmbeddingModel', 'MistralModel', 'LlavaLlamaForCausalLM', 'LlavaQwenForCausalLM', 'LlavaMistralForCausalLM', 'LlavaVidForCausalLM', 'MiniCPMForCausalLM', 'MistralForCausalLM', 'MixtralForCausalLM', 'QuantMixtralForCausalLM', 'QWenLMHeadModel', 'Qwen2ForCausalLM', 'Qwen2MoeForCausalLM', 'StableLmForCausalLM', 'YiVLForCausalLM']
vllm 0.5.5 supports this architecture. why has sglang always been unsupported?
Reproduction
1
Environment
1