sgl-project/sglang · critical · ValueError

Weight {name} not found in params_dict

Error message

Weight {name} not found in params_dict

What it means

Identical guard to other multimodal loaders: while loading the vision tower of Step3p7, every vision checkpoint key must match a named parameter of the model or loading fails.

Source

Thrown at python/sglang/srt/models/step3p7.py:190

            if "vision_model" in name or "vit_large_projector" in name:
                # Strip leading "model." for vision weights (NVFP4 format)
                if name.startswith("model."):
                    name = name[len("model.") :]
                name = name.replace(r".attn.in_proj_weight", r".attn.qkv_proj.weight")
                name = name.replace(r".attn.in_proj_bias", r".attn.qkv_proj.bias")
                name = name.replace(r".attn.out_proj.bias", r".attn.proj.bias")
                name = name.replace(r".attn.out_proj.weight", r".attn.proj.weight")
                name = name.replace(".mlp.c_fc", ".mlp.fc1")
                name = name.replace(".mlp.c_proj", ".mlp.fc2")
                vision_weights.append((name, loaded_weight))
            else:
                language_weights.append((name, loaded_weight))

        # Load vision tower weights
        params_dict = dict(self.named_parameters(remove_duplicate=False))
        for name, loaded_weight in vision_weights:
            if name not in params_dict:
                raise ValueError(f"Weight {name} not found in params_dict")
            param = params_dict[name]
            weight_loader = getattr(param, "weight_loader", default_weight_loader)
            weight_loader(param, loaded_weight)

        # Load language model weights
        if language_weights:
            self.language_model.load_weights(language_weights)


EntryClass = Step3p7ForConditionalGeneration

View on GitHub (pinned to 0132848349)

Solutions

1. Diff checkpoint vision keys vs model named_parameters and fix naming/config
2. Confirm vision_config in config.json matches the trained model
3. Filter or remap known-renamed vision weights in load_weights

Example fix

# debug
ck = set(n for n,_ in vision_weights)
mp = set(dict(self.named_parameters(remove_duplicate=False)))
print(sorted(ck - mp))

missing = {n for n,_ in vision_weights} - set(dict(model.named_parameters(remove_duplicate=False)))
if missing: print('unmapped:', sorted(missing)[:10])

When it happens

Trigger: Step3p7 checkpoint vision keys that don't match model parameter names (renamed modules, different vision_config, extra keys).

Common situations: Checkpoint/model revision skew; vision config fields (hidden size, num heads) producing differently-named or absent modules.

Related errors

• Weight {name} not found in params_dict
• Incomplete Diffusers H3 fused parameters: {incomplete}
• qkv weight has incompatible output dim for grouped checkpoin
• Some weights are not initialized from checkpoints: {unloaded
• Only 1 nextn layer is supported for Step3p5 checkpoints.

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/f7892604b2070bea. Report an issue: GitHub.