Blame for eval_joint.py

"""eval_joint.py — stock vs JOINT (LLM-LoRA + full head, fullv3) on the identical 9-dialogue set.
Baked fixes: demo-exact generate, sf tuple unpacks, save_audio write path, safe drift
aggregation, Whisper return_timestamps=True, adapter download from model repo via
list_repo_files with shutil.copyfile (HF cache paths are RELATIVE symlinks — os.replace
leaves dangling links), load_lora_assets (imported standalone from vibevoice-community
main with sys.modules registration) with component asserts.
Env: ARMS (default "stock,joint"). Pushes to ACloudCenter/vibevoice-eval-jointv1. Prints EVAL DONE.
"""
import importlib.util, json, os, shutil, sys, time
sys.path.insert(0, "/VibeVoice-classic")

import numpy as np
import soundfile as sf
import torch
import jiwer
from huggingface_hub import hf_hub_download, HfApi
from resemblyzer import VoiceEncoder, preprocess_wav

from vibevoice.modular.modeling_vibevoice_inference import VibeVoiceForConditionalGenerationInference
from vibevoice.processor.vibevoice_processor import VibeVoiceProcessor

_spec = importlib.util.spec_from_file_location(
    "vibevoice.modular.lora_loading", "/vv-main/vibevoice/modular/lora_loading.py")
assert _spec is not None and _spec.loader is not None, "lora_loading.py not found at /vv-main"
lora_loading = importlib.util.module_from_spec(_spec)
sys.modules[_spec.name] = lora_loading  # dataclass machinery needs the module registered
_spec.loader.exec_module(lora_loading)
load_lora_assets = lora_loading.load_lora_assets

MODEL = "microsoft/VibeVoice-1.5B"
ADAPTER_REPO = "ACloudCenter/vibevoice-lora-joint"
ADAPTER_SUB = "fullv3/lora"
OUT_REPO = "ACloudCenter/vibevoice-eval-jointv1"
SR = 24000
device = "cuda"
ARMS = [a.strip() for a in os.environ.get("ARMS", "stock,joint").split(",")]

SHORTS = [
    "Speaker 1: Did you see the meteor shower last night?\nSpeaker 2: I missed it! Was it worth staying up for?\nSpeaker 1: Absolutely. The sky lit up green for a full minute.\nSpeaker 2: Okay, now I'm actually upset I slept through it.",
    "Speaker 1: The bread machine finally arrived.\nSpeaker 2: So you're a baker now?\nSpeaker 1: I'm a disaster with a bread machine. The first loaf came out like a brick.\nSpeaker 2: Send me a photo. Immediately.",
    "Speaker 1: My flight lands at six, so don't wait for dinner.\nSpeaker 2: I'll keep something warm anyway.\nSpeaker 1: You really don't have to.\nSpeaker 2: I know. That's why I'm doing it.",
]
MEDIUMS = [
    "Speaker 1: I've been reading about how old lighthouse keepers kept journals.\nSpeaker 2: What did they write about?\nSpeaker 1: Mostly weather, ships passing, and how lonely it was.\nSpeaker 2: That sounds peaceful and terrible at the same time.\nSpeaker 1: One keeper wrote that the light was his only conversation for months.\nSpeaker 2: Honestly, that might fix my screen time problem.\nSpeaker 1: Or break you completely. There was no off switch on the ocean.",
    "Speaker 1: The city is redoing the bike lane on Fifth Street.\nSpeaker 2: Finally. That stretch is dangerous.\nSpeaker 1: They're adding a physical barrier between cars and cyclists.\nSpeaker 2: About time. I've almost been clipped twice this month.\nSpeaker 1: Construction starts Monday, so expect detours.\nSpeaker 2: Worth it if it means I stop white-knuckling my handlebars.",
    "Speaker 1: My grandmother used to make bread every Sunday morning.\nSpeaker 2: Sourdough?\nSpeaker 1: A dense rye. She said the recipe came from her grandmother.\nSpeaker 2: Four generations of the same loaf.\nSpeaker 1: I still have her mixing bowl. It's older than the house.\nSpeaker 2: That bowl has earned a museum display.",
]
LONGS = [
    "Speaker 1: Let's talk about sleep. Everyone says eight hours, but nobody agrees on anything else.\nSpeaker 2: Right, and half the advice contradicts the other half.\nSpeaker 1: Some people swear by cold rooms. Others say the temperature barely matters.\nSpeaker 2: The one thing that's consistent is the schedule. Same bedtime, same wake time.\nSpeaker 1: Even on weekends, which is where everyone cheats.\nSpeaker 2: That's exactly where I cheat.\nSpeaker 1: The researchers say your body doesn't know it's Saturday.\nSpeaker 2: My body absolutely knows it's Saturday.\nSpeaker 1: Apparently not, biologically speaking.\nSpeaker 2: Fine, I'll set an alarm for Sunday too. This conversation is costing me my weekend.\nSpeaker 1: Sleep hygiene comes with sacrifices.\nSpeaker 2: Don't lecture me at nine on a Sunday morning.",
    "Speaker 1: The bakery on the corner closed last week.\nSpeaker 2: No! That place had the best almond croissants in the city.\nSpeaker 1: Rent tripled. The owner said she couldn't make the numbers work anymore.\nSpeaker 2: Where does a neighborhood bakery even go from there?\nSpeaker 1: She's moving production to a shared kitchen across the river and doing delivery only.\nSpeaker 2: So no more counter, no more coffee, just a website?\nSpeaker 1: Just a website. She seemed okay about it, which surprised me.\nSpeaker 2: Thirty years behind that counter and she seemed okay?\nSpeaker 1: Maybe she's more practical than sentimental.\nSpeaker 2: I'm going to miss walking past that smell every morning.\nSpeaker 1: Delivery still smells. It just doesn't linger.",
    "Speaker 1: I finally fixed the shed door.\nSpeaker 2: The one that's been hanging crooked since spring?\nSpeaker 1: Since spring. The hinges were stripped, not the frame, which saved me a full rebuild.\nSpeaker 2: How long did that take?\nSpeaker 1: Four hours, two of which were finding screws that actually fit.\nSpeaker 2: The hardware store run is always the real project.\nSpeaker 1: Two trips. One for the wrong screws, one for the right ones.\nSpeaker 2: Classic. Did you paint it?\nSpeaker 1: Not yet. The wood needs to dry out first, apparently.\nSpeaker 2: So the project continues.\nSpeaker 1: The project continues. But the door closes now, which is more than March could say.",
]


def download_adapter(dest="/adapter"):
    api = HfApi()
    files = [f for f in api.list_repo_files(ADAPTER_REPO) if f.startswith(ADAPTER_SUB + "/")]
    assert len(files) >= 4, f"adapter files not found: {files}"
    for f in files:
        local = hf_hub_download(ADAPTER_REPO, f)
        name = f[len(ADAPTER_SUB) + 1:]
        op = os.path.join(dest, name)
        os.makedirs(os.path.dirname(op), exist_ok=True)
        shutil.copyfile(local, op)  # NOT os.replace: cache paths are relative symlinks
    listing = sorted(os.listdir(dest))
    print(f"ADAPTER OK: {listing}", flush=True)
    # follow-the-link verification: os.listdir can show dangling symlinks
    for must in ("adapter_config.json", "adapter_model.safetensors", "diffusion_head_full.bin"):
        assert os.path.isfile(os.path.join(dest, must)), f"dangling or missing: {must}"
    return dest


def load_model(arm):
    model = VibeVoiceForConditionalGenerationInference.from_pretrained(
        MODEL, torch_dtype=torch.bfloat16, low_cpu_mem_usage=True,
        attn_implementation="sdpa").to(device).eval()
    if arm == "joint":
        report = load_lora_assets(model, "/adapter")
        loaded = [n for n, ok in (("language_model", report.language_model),
                                  ("diffusion_head_lora", report.diffusion_head_lora),
                                  ("diffusion_head_full", report.diffusion_head_full),
                                  ("acoustic_connector", report.acoustic_connector),
                                  ("semantic_connector", report.semantic_connector)) if ok]
        print(f"JOINT LOADED: {loaded}", flush=True)
        assert "language_model" in loaded and "diffusion_head_full" in loaded, loaded
    return model


def gt_text(script):
    return " ".join(line.split(": ", 1)[1].strip() for line in script.strip().splitlines())


def main():
    api = HfApi()
    download_adapter()
    scripts = ([("short", i, t) for i, t in enumerate(SHORTS)]
               + [("medium", i, t) for i, t in enumerate(MEDIUMS)]
               + [("long", i, t) for i, t in enumerate(LONGS)])

    os.makedirs("/ref", exist_ok=True)
    ref_wavs = {}
    for spk, name in ((1, "en-Carter_man.wav"), (2, "en-Maya_woman.wav")):
        wav, sr = sf.read(f"/VibeVoice-classic/demo/voices/{name}")
        assert sr == SR
        ref_wavs[spk] = wav.astype(np.float32)
        sf.write(f"/ref/s{spk}.wav", ref_wavs[spk], SR)
    ref_paths = {spk: f"/ref/s{spk}.wav" for spk in (1, 2)}

    proc = VibeVoiceProcessor.from_pretrained(MODEL)
    from transformers import pipeline
    transcriber = pipeline("automatic-speech-recognition", model="openai/whisper-large-v3",
                           device=device, torch_dtype=torch.float16)
    encoder = VoiceEncoder(device)
    ref_wav = np.concatenate([ref_wavs[1], np.zeros(SR // 4, dtype=np.float32), ref_wavs[2]])
    emb_ref = encoder.embed_utterance(preprocess_wav(ref_wav, SR))

    results = []
    os.makedirs("/out", exist_ok=True)
    for arm in ARMS:
        print(f"=== ARM {arm} ===", flush=True)
        model = load_model(arm)
        model.set_ddpm_inference_steps(num_steps=10)
        for tier, idx, script in scripts:
            t0 = time.time()
            with torch.no_grad():
                inputs = proc(text=[script], voice_samples=[[ref_paths[1], ref_paths[2]]],
                              padding=True, return_tensors="pt", return_attention_mask=True)
                out = model.generate(
                    **inputs, max_new_tokens=None, cfg_scale=1.3,
                    tokenizer=proc.tokenizer, generation_config={"do_sample": False})
            gen_s = time.time() - t0
            fn = f"{arm}_{tier}_{idx}.wav"
            proc.save_audio(out.speech_outputs[0], output_path=f"/out/{fn}")
            wav = sf.read(f"/out/{fn}", dtype="float32")[0]
            wav = np.asarray(wav, dtype=np.float32)
            audio_s = len(wav) / SR
            rtf = gen_s / max(audio_s, 1e-6)
            hyp = transcriber(f"/out/{fn}", generate_kwargs={"language": "english"},
                              return_timestamps=True)["text"].strip()
            gt = gt_text(script)
            wer = float(jiwer.wer(gt, hyp))
            emb_gen = encoder.embed_utterance(preprocess_wav(f"/out/{fn}"))
            sim = float(np.dot(emb_gen, emb_ref) / (np.linalg.norm(emb_gen) * np.linalg.norm(emb_ref)))
            drift = None
            if tier == "long" and audio_s > 60:
                e1 = encoder.embed_utterance(preprocess_wav(wav[: 30 * SR], SR))
                e2 = encoder.embed_utterance(preprocess_wav(wav[-30 * SR:], SR))
                n = np.linalg.norm(e1) * np.linalg.norm(e2)
                drift = float(np.dot(e1, e2) / n) if n > 0 else None
            results.append({"arm": arm, "tier": tier, "idx": idx, "audio_s": round(audio_s, 1),
                            "rtf": round(rtf, 3), "wer": round(wer, 3), "sim": round(sim, 3),
                            "drift": round(drift, 3) if drift is not None else None})
            api.upload_file(path_or_fileobj=f"/out/{fn}", path_in_repo=fn, repo_id=OUT_REPO)
            print(f"{arm} {tier}{idx}: rtf {rtf:.3f} wer {wer:.3f} sim {sim:.3f} "
                  f"dur {audio_s:.1f}s drift {drift}", flush=True)
        del model
        torch.cuda.empty_cache()

    import statistics as stat
    def agg(arm, key):
        vals = [r[key] for r in results if r["arm"] == arm and r[key] is not None]
        return round(stat.mean(vals), 3) if vals else None
    summary = {arm: {"wer": agg(arm, "wer"), "sim": agg(arm, "sim"),
                     "rtf": agg(arm, "rtf"), "drift": agg(arm, "drift")} for arm in ARMS}
    blob = {"summary": summary, "results": results}
    with open("/out/results.json", "w") as f:
        json.dump(blob, f, indent=1)
    api.upload_file(path_or_fileobj="/out/results.json", path_in_repo="results.json", repo_id=OUT_REPO)
    print("SUMMARY:", json.dumps(summary), flush=True)
    print("EVAL DONE", flush=True)


if __name__ == "__main__":
    main()