diff --git a/docs/docs/hardware-platforms/tpu.mdx b/docs/docs/hardware-platforms/tpu.mdx index b3d7f2516..f386ebf9c 100644 --- a/docs/docs/hardware-platforms/tpu.mdx +++ b/docs/docs/hardware-platforms/tpu.mdx @@ -165,7 +165,7 @@ SGLang-JAX provides comprehensive TPU-optimized features for production LLM serv ## Optimized Model List -The following models have been tested and optimized for TPU deployment: +SGLang-JAX supports the following model families. Use the linked TPU cookbooks for deployment instructions.
| Model Family | -Performance Status | +Cookbook | |
|---|---|---|---|
| Qwen 3 | -⭐ Recommended for production | +Qwen3 | |
| Qwen 3 MoE | -⭐ Best performance | +Qwen3 MoE | |
| Qwen 2 | -Needs improvement | -||
| Qwen 2 MoE | -Needs improvement | -||
| Qwen 1.5 | -Needs improvement | +Qwen2.5-VL | +Qwen2.5-VL |
| Llama/LLaMA | -Needs improvement | +Llama 3.1 / Llama 3.3 70B | +|
| DeepSeek V3 / R1 | +DeepSeek V3 / DeepSeek R1 | +||
| Kimi-Linear | +Kimi-Linear | ||
| Grok-2 | -Needs improvement | +Grok-2 | |
| Gemma 2 | -Verified on TPU | +Gemma 2 | |
| Bailing MoE | -Needs improvement | +MiMo-V2-Flash | +MiMo-V2-Flash | +
| MiMo-V2.5-Pro | +MiMo-V2.5-Pro | +||
| Ling-2.6 | +Ling-2.6 |