sgl-project/sglang · error · ValueError
Unsupported activation: {hidden_act}. Only silu is supported
Error message
Unsupported activation: {hidden_act}. Only silu is supported for now. What it means
Solar's MLP hardcodes SiluAndMul and rejects any other hidden_act from config. The fused activation kernel only implements silu, so a config specifying gelu/pytorch_silu/etc. cannot be served correctly and init fails.
Source
Thrown at python/sglang/srt/models/solar.py:89
prefix: str = "",
) -> None:
super().__init__()
self.gate_up_proj = MergedColumnParallelLinear(
input_size=hidden_size,
output_sizes=[intermediate_size] * 2,
bias=bias,
quant_config=quant_config,
prefix=f"{prefix}.gate_up_proj",
)
self.down_proj = RowParallelLinear(
input_size=intermediate_size,
output_size=hidden_size,
bias=bias,
quant_config=quant_config,
prefix=f"{prefix}.down_proj",
)
if hidden_act != "silu":
raise ValueError(
f"Unsupported activation: {hidden_act}. "
"Only silu is supported for now."
)
self.act_fn = SiluAndMul()
def forward(self, x):
gate_up, _ = self.gate_up_proj(x)
x = self.act_fn(gate_up)
x, _ = self.down_proj(x)
return x
class SolarAttention(nn.Module):
def __init__(
self,
config: PretrainedConfig,
hidden_size: int,View on GitHub (pinned to 0132848349)
Solutions
- Edit the model's config.json hidden_act back to "silu" if the checkpoint was trained with silu
- If the model genuinely uses another activation, use a backend that supports it (e.g. HF transformers) or add support in the MLP
- Re-download the original checkpoint config from the model repo
Example fix
// config.json // before: "hidden_act": "gelu" // after: "hidden_act": "silu"
Defensive patterns
Strategy: validation
Validate before calling
assert config.hidden_act in ("silu", "swish"), config.hidden_act Prevention
- Validate activation string against model support before launching
When it happens
Trigger: Loading a Solar-family checkpoint whose config.json hidden_act (or intermediate_act_fn) is anything other than 'silu'.
Common situations: Fine-tuned/merged Solar variants saved with a different activation string, or quantization/conversion tooling that rewrote config.json activation fields.
Related errors
- Unsupported activation type: {act_type}
- Activation function {act_fn_name!r} is not supported.
- intermediate_size must be specified for scaled activation fu
- Unsupported activation: {hidden_act}. Only silu is supported
- This browser cannot encode H.264 MP4
AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28).
Data as JSON: /api/errors/319b851af3f931d1.
Report an issue: GitHub.