What causes "Invalid wtype: %s, using default\n" in mudler/LocalAI?

Use an exact sd_type name from the vendored str_to_sd_type (e.g. f16, q8_0, q4_0)

Invalid wtype: %s, using default\n

Error message

Invalid wtype: %s, using default\n

What it means

Warning from the option parser: 'wtype=<value>' (weight quantization type, e.g. f16/q8_0/q4_0) is not recognized by str_to_sd_type(), so the model's own default weight type is used (ctx_params.wtype is only assigned when the parse succeeded). Load proceeds; the model may use a different precision than intended, affecting VRAM and quality.

Source

Thrown at backend/go/stablediffusion-ggml/cpp/gosd.cpp:577

                fprintf(stderr, "Invalid prediction: %s, using default\n", optval);
            }
        }
        if (!strcmp(optname, "lora_apply_mode")) {
            lora_apply_mode_t parsed = str_to_lora_apply_mode(optval);
            if (parsed != LORA_APPLY_MODE_COUNT) {
                lora_apply_mode = parsed;
                fprintf(stderr, "Found lora_apply_mode: %s\n", optval);
            } else {
                fprintf(stderr, "Invalid lora_apply_mode: %s, using default\n", optval);
            }
        }
        if (!strcmp(optname, "wtype")) {
            sd_type_t parsed = str_to_sd_type(optval);
            if (parsed != SD_TYPE_COUNT) {
                wtype = parsed;
                fprintf(stderr, "Found wtype: %s\n", optval);
            } else {
                fprintf(stderr, "Invalid wtype: %s, using default\n", optval);
            }
        }
    }

    fprintf(stderr, "parsed options\n");

    // Build embeddings vector from directory if provided
    build_embedding_vec(embedding_dir);

    fprintf (stderr, "Creating context\n");
    sd_ctx_params_init(&ctx_params);
    ctx_params.model_path = model;
    ctx_params.clip_l_path = clip_l_path;
    ctx_params.clip_g_path = clip_g_path;
    ctx_params.clip_vision_path = clip_vision_path;
    ctx_params.t5xxl_path = t5xxl_path;
    ctx_params.llm_path = llm_path;
    ctx_params.llm_vision_path = llm_vision_path;

View on GitHub (pinned to 44413a9d06)

Solutions

Use an exact sd_type name from the vendored str_to_sd_type (e.g. f16, q8_0, q4_0)
Omit wtype to accept the model default precision
Watch for the 'Found wtype:' confirmation line to verify it took effect

Example fix

# before
options: "...,wtype=Q8"  # not a valid sd type string

# after
options: "...,wtype=q8_0"

Defensive patterns

Strategy: validation

Validate before calling

// Go caller: whitelist weight types
validWtype := map[string]bool{"f32": true, "f16": true, "q8_0": true, "q4_0": true} // match vendored names
if o, ok := opts.Raw["wtype"]; ok && !validWtype[o] {
    return fmt.Errorf("invalid wtype %q", o)
}

Prevention

Use sd.cpp weight-type spellings, not llama.cpp quant abbreviations
Verify quantization actually applied by checking the 'Found wtype:' line and memory usage

When it happens

Trigger: Passing a wtype string not supported by the vendored stable-diffusion.cpp (unsupported quant name, typo, wrong case).

Common situations: Trying to shrink a large model with a quant name that this build does not implement; copying ggml quant names from llama.cpp that sd.cpp spells differently.

Related errors

AI-assisted analysis of mudler/LocalAI@44413a9d06 (2026-08-15). Data as JSON: /api/errors/47a1539149070b43. Report an issue: GitHub.