sgl-project/sglang · error · ValueError

Length mismatch: lora_nicknames has {len(lora_nicknames)} it

Error message

Length mismatch: lora_nicknames has {len(lora_nicknames)} items, but strengths has {len(strengths)} items

What it means

Same per-adapter list validation in _apply_lora_to_layers, but for strengths: the strengths list must match lora_nicknames in length so each adapter has a scaling factor.

Source

Thrown at python/sglang/multimodal_gen/runtime/pipelines_core/lora/pipeline.py:649

        Args:
            lora_layers: The dictionary of LoRA layers to apply weights to.
            lora_nicknames: The list of nicknames of the LoRA adapters.
            lora_paths: The list of paths to the LoRA adapters. Must match length of lora_nicknames.
            rank: The distributed rank (for logging).
            strengths: The list of LoRA strengths for merge. Must match length of lora_nicknames.
            clear_existing: If True, clear existing LoRA weights before adding new ones.

        Returns:
            The number of layers that had LoRA weights applied.
        """
        if len(lora_paths) != len(lora_nicknames):
            raise ValueError(
                f"Length mismatch: lora_nicknames has {len(lora_nicknames)} items, "
                f"but lora_paths has {len(lora_paths)} items"
            )
        if len(strengths) != len(lora_nicknames):
            raise ValueError(
                f"Length mismatch: lora_nicknames has {len(lora_nicknames)} items, "
                f"but strengths has {len(strengths)} items"
            )

        adapted_count = 0
        missing_layers_by_adapter = [[] for _ in lora_nicknames]
        applied_count_by_adapter = [0 for _ in lora_nicknames]
        for name, layer in lora_layers.items():
            # Apply all LoRA adapters in order
            for idx, (nickname, path, lora_strength) in enumerate(
                zip(lora_nicknames, lora_paths, strengths)
            ):
                lora_A_name = name + ".lora_A"
                lora_B_name = name + ".lora_B"
                if (
                    lora_A_name in self.lora_adapters[nickname]
                    and lora_B_name in self.lora_adapters[nickname]
                ):

View on GitHub (pinned to 0132848349)

Solutions

1. Provide one strength per nickname, e.g. strengths=[1.0]*len(lora_nicknames)
2. Omit strengths entirely if defaults are acceptable

Example fix

# before
pipeline.set_lora(lora_paths=paths, lora_nicknames=['a','b'], strengths=[0.8])
# after
pipeline.set_lora(lora_paths=paths, lora_nicknames=['a','b'], strengths=[0.8, 0.8])

if strengths is not None: assert len(strengths) == len(lora_nicknames)

When it happens

Trigger: Calling set_lora(lora_nicknames=['a','b'], strengths=[1.0]) — fewer (or more) strength values than adapters.

Common situations: Passing a single scalar strength when multiple adapters are composed; strengths list built for an older adapter set after adding a new adapter.

Related errors

• Length mismatch: lora_nicknames has {len(lora_nicknames)} it
• g and beta must cover every q token
• attn_sink requires topk_length to be provided as well
• missing value for {a} (expected e.g. `{a} 2,4`)
• combined_history=True requires direction=0 (bidi)

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/365fedebd9ac370b. Report an issue: GitHub.