mudler/LocalAI · critical

3

3

Error message

[omnivoice-cpp] FATAL: ov_init failed: %s\n

What it means

ov_init() returned NULL after params were filled (model_path, codec_path, use_fa, clamp_fp16). This is the engine-level load failure (return code 3); unlike moss-tts, the OmniVoice engine exposes the reason through ov_last_error(), which the message prints. Causes range from unreadable files to flash-attention (use_fa) or fp16 options unsupported on the current backend.

Source

Thrown at backend/go/omnivoice-cpp/cpp/gomnivoicecpp.cpp:53

    }
    if (!codec_path || codec_path[0] == '\0') {
        fprintf(stderr, "[omnivoice-cpp] ERROR: codec_path is required\n");
        return 2;
    }

    ov_init_params p;
    ov_init_default_params(&p);
    p.model_path = model_path;
    p.codec_path = codec_path;
    p.use_fa = use_fa != 0;
    p.clamp_fp16 = clamp_fp16 != 0;

    fprintf(stderr, "[omnivoice-cpp] Loading model=%s codec=%s\n", model_path,
            codec_path);

    g_ctx = ov_init(&p);
    if (!g_ctx) {
        fprintf(stderr, "[omnivoice-cpp] FATAL: ov_init failed: %s\n",
                ov_last_error());
        return 3;
    }
    fprintf(stderr, "[omnivoice-cpp] Model loaded (%s)\n", ov_version());
    return 0;
}

// Fill an ov_tts_params from the flat wrapper arguments.
static void fill_params(ov_tts_params *tp, const char *text, const char *lang,
                        const char *instruct, const float *ref_samples,
                        int ref_n, const char *ref_text, long long seed,
                        int denoise) {
    ov_tts_default_params(tp);
    tp->text = text ? text : "";
    tp->lang = lang ? lang : "";
    if (instruct && instruct[0] != '\0')
        tp->instruct = instruct;
    if (ref_samples && ref_n > 0) {

View on GitHub (pinned to 44413a9d06)

Solutions

1. Read the ov_last_error() text in the FATAL line — it names the exact cause; fix that first.
2. Retry with use_fa=0 if the error mentions flash attention.
3. Verify model and codec files exist, match the same release, and have expected sizes.
4. Disable clamp_fp16 or free memory if the error indicates allocation/precision problems.

Example fix

// before
rc := C.omni_load(m, c, 1, 1) // fa + fp16 clamp on unsupported backend
// after
rc := C.omni_load(m, c, 0, 0) // conservative flags first, then re-enable

for _, p := range []string{modelPath, codecPath} {
    if _, err := os.Stat(p); err != nil {
        return fmt.Errorf("omnivoice artifact missing: %w", err)
    }
}

When it happens

Trigger: ov_init failing on missing/corrupt model or codec files, incompatible quantization, use_fa=1 on a backend without flash-attention support, or clamp_fp16 interactions with an fp16-incompatible device; also plain OOM during weight load.

Common situations: use_fa enabled on CPU-only or older GPU builds; model/codec from different OmniVoice releases; truncated downloads; memory-constrained containers.

Related errors

• 1
• [moss-tts-cpp] ERROR: local_path is required\n
• [moss-tts-cpp] FATAL: moss_local_load failed\n
• 2
• [omnivoice-cpp] ERROR: model not loaded\n

AI-assisted analysis of mudler/LocalAI@44413a9d06 (2026-08-15). Data as JSON: /api/errors/81c815e5c63e2d58. Report an issue: GitHub.