feat: LoRA fine-tuning for SelVA generator

Teaches the model new/partial sound classes from custom video+audio pairs.
Only ~10 MB of adapter weights are trained vs ~4.4 GB for the full model.

selva_core/model/lora.py
  LoRALinear: wraps nn.Linear with frozen base + trainable A/B matrices.
  B initialised to zero → zero adapter contribution at init.
  apply_lora(): walks named_modules, replaces matching nn.Linear in-place.
  Default target: "attn.qkv" (all 21 SelfAttention QKV projections in
  large_44k). Add "linear1" to also wrap post-attention output projections.
  get_lora_state_dict() / load_lora() for ~10 MB save/load.

train_lora.py (standalone script, no ComfyUI dependency)
  Data format: directory of video files + optional prompts.txt
  ("filename: description"). Falls back to directory name as prompt.
  Pre-extracts features for all clips into RAM, then trains from those.
  Training loop: encode audio→latent (need_vae_encoder=True), flow
  matching MSE loss on velocity prediction, backward on LoRA params only.
  Saves adapter_stepNNNNN.pt checkpoints + adapter_final.pt with metadata.
  Key verified interfaces used:
    encode_audio() → DiagonalGaussianDistribution; .mode().clone() required
    normalize() is in-place
    forward(latent, clip_f, sync_f, text_f, t) takes raw tensors

nodes/selva_lora_loader.py (SelVA LoRA Loader ComfyUI node)
  Loads .pt adapter, deep-copies the generator, applies LoRA, loads weights.
  strength param scales lora_B to adjust adapter contribution at inference.
  Reads rank/alpha/target from embedded metadata if present.
  Returns a patched SELVA_MODEL bundle for use with the existing Sampler.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
2026-04-05 14:38:46 +02:00
parent b519b042e2
commit 437c62b28f
4 changed files with 629 additions and 0 deletions
+1
View File
@@ -5,6 +5,7 @@ _NODES = {
"SelvaModelLoader": (".selva_model_loader", "SelvaModelLoader", "SelVA Model Loader"), "SelvaModelLoader": (".selva_model_loader", "SelvaModelLoader", "SelVA Model Loader"),
"SelvaFeatureExtractor": (".selva_feature_extractor", "SelvaFeatureExtractor", "SelVA Feature Extractor"), "SelvaFeatureExtractor": (".selva_feature_extractor", "SelvaFeatureExtractor", "SelVA Feature Extractor"),
"SelvaSampler": (".selva_sampler", "SelvaSampler", "SelVA Sampler"), "SelvaSampler": (".selva_sampler", "SelvaSampler", "SelVA Sampler"),
"SelvaLoraLoader": (".selva_lora_loader", "SelvaLoraLoader", "SelVA LoRA Loader"),
} }
for key, (module_path, class_name, display_name) in _NODES.items(): for key, (module_path, class_name, display_name) in _NODES.items():
+93
View File
@@ -0,0 +1,93 @@
import copy
import torch
import folder_paths
from .utils import SELVA_CATEGORY
from selva_core.model.lora import apply_lora, load_lora
class SelvaLoraLoader:
@classmethod
def INPUT_TYPES(cls):
return {
"required": {
"model": ("SELVA_MODEL",),
"adapter_path": ("STRING", {
"default": "",
"tooltip": "Path to a LoRA adapter .pt file produced by train_lora.py.",
}),
"strength": ("FLOAT", {
"default": 1.0, "min": 0.0, "max": 2.0, "step": 0.05,
"tooltip": "Scale applied to all LoRA contributions. "
"1.0 = full adapter strength. "
"0.0 = effectively disables the adapter. "
"Values above 1.0 exaggerate the effect.",
}),
},
}
RETURN_TYPES = ("SELVA_MODEL",)
RETURN_NAMES = ("model",)
OUTPUT_TOOLTIPS = ("Model with LoRA adapter applied — connect to Sampler.",)
FUNCTION = "load"
CATEGORY = SELVA_CATEGORY
DESCRIPTION = (
"Loads a LoRA adapter produced by train_lora.py and applies it to the generator. "
"The base model is not modified — a shallow copy of the model bundle is returned."
)
def load(self, model: dict, adapter_path: str, strength: float) -> tuple:
if not adapter_path.strip():
raise ValueError("[SelVA LoRA] adapter_path is empty.")
# Resolve path: allow absolute or relative to ComfyUI base
from pathlib import Path
p = Path(adapter_path)
if not p.is_absolute():
p = Path(folder_paths.base_path) / p
if not p.exists():
raise FileNotFoundError(f"[SelVA LoRA] Adapter not found: {p}")
checkpoint = torch.load(str(p), map_location="cpu", weights_only=False)
# Support both raw state_dict and {state_dict, meta} formats
if isinstance(checkpoint, dict) and "state_dict" in checkpoint:
state_dict = checkpoint["state_dict"]
meta = checkpoint.get("meta", {})
else:
state_dict = checkpoint
meta = {}
rank = int(meta.get("rank", 16))
alpha = float(meta.get("alpha", float(rank)))
target = list(meta.get("target", ["attn.qkv"]))
print(f"[SelVA LoRA] Loading adapter: {p.name}", flush=True)
print(f"[SelVA LoRA] rank={rank} alpha={alpha} target={target} strength={strength}",
flush=True)
# Shallow-copy the model bundle so the original generator is not mutated
patched = {**model}
generator = copy.deepcopy(model["generator"])
n = apply_lora(generator, rank=rank, alpha=alpha, target_suffixes=tuple(target))
if n == 0:
raise RuntimeError(
f"[SelVA LoRA] No layers matched target={target}. "
"Check that the adapter was trained with the same target suffixes."
)
load_lora(generator, state_dict)
# Apply strength scaling: multiply all lora_B params by strength
# (lora_B is initialised to zero, so scaling A is equivalent but less clean)
if strength != 1.0:
with torch.no_grad():
for name, param in generator.named_parameters():
if "lora_B" in name:
param.mul_(strength)
generator.to(model["generator"].parameters().__next__().device)
patched["generator"] = generator
print(f"[SelVA LoRA] Applied {n} LoRA layers.", flush=True)
return (patched,)
+116
View File
@@ -0,0 +1,116 @@
"""
LoRA (Low-Rank Adaptation) for SelVA / MMAudio generator.
Usage:
from selva_core.model.lora import apply_lora, get_lora_state_dict, load_lora
n = apply_lora(net_generator, rank=16, alpha=16.0)
print(f"Wrapped {n} linear layers with LoRA")
# ... train only LoRA params ...
torch.save(get_lora_state_dict(net_generator), "adapter.pt")
# Later, at inference:
apply_lora(net_generator, rank=16, alpha=16.0)
load_lora(net_generator, torch.load("adapter.pt"))
"""
import math
import torch
import torch.nn as nn
class LoRALinear(nn.Module):
"""nn.Linear with a frozen base weight and trainable low-rank A/B matrices.
Output: base(x) + (x @ A.T @ B.T) * (alpha / rank)
A is initialised with Kaiming uniform; B is initialised to zero so the
adapter contribution starts at zero and does not disturb pretrained behaviour.
"""
def __init__(self, linear: nn.Linear, rank: int, alpha: float):
super().__init__()
in_f = linear.in_features
out_f = linear.out_features
self.linear = linear
linear.weight.requires_grad_(False)
if linear.bias is not None:
linear.bias.requires_grad_(False)
self.lora_A = nn.Parameter(torch.empty(rank, in_f))
self.lora_B = nn.Parameter(torch.zeros(out_f, rank))
self.scale = alpha / rank
nn.init.kaiming_uniform_(self.lora_A, a=math.sqrt(5))
def forward(self, x: torch.Tensor) -> torch.Tensor:
return self.linear(x) + (x @ self.lora_A.T @ self.lora_B.T) * self.scale
def extra_repr(self) -> str:
rank = self.lora_A.shape[0]
return (f"in={self.linear.in_features}, out={self.linear.out_features}, "
f"rank={rank}, scale={self.scale:.4f}")
def apply_lora(
model: nn.Module,
rank: int = 16,
alpha: float = None,
target_suffixes: tuple = ("attn.qkv",),
) -> int:
"""Replace matching nn.Linear layers with LoRALinear in-place.
Args:
model: The module to modify (typically net_generator).
rank: LoRA rank.
alpha: LoRA alpha (scaling). Defaults to rank (scale = 1.0).
target_suffixes: Tuple of module name suffixes to wrap. Default is
("attn.qkv",) which targets all SelfAttention QKV
projections in the MM-DiT generator.
Add "linear1" to also wrap post-attention output projections.
Returns:
Number of linear layers wrapped.
"""
if alpha is None:
alpha = float(rank)
count = 0
for name, module in list(model.named_modules()):
if not any(name.endswith(s) for s in target_suffixes):
continue
if not isinstance(module, nn.Linear):
continue
parts = name.split(".")
parent = model
for part in parts[:-1]:
parent = getattr(parent, part)
setattr(parent, parts[-1], LoRALinear(module, rank, alpha))
count += 1
return count
def get_lora_state_dict(model: nn.Module) -> dict:
"""Return a state dict containing only LoRA parameters (lora_A and lora_B)."""
return {k: v for k, v in model.state_dict().items() if "lora_" in k}
def load_lora(model: nn.Module, state_dict: dict) -> None:
"""Load LoRA weights into a model that has already had apply_lora() called.
Non-LoRA keys in state_dict are ignored (strict=False). Non-LoRA model
parameters are not modified.
"""
missing, unexpected = model.load_state_dict(state_dict, strict=False)
bad = [k for k in unexpected if "lora_" not in k]
if bad:
print(f"[LoRA] Warning: unexpected non-LoRA keys ignored: {bad}")
lora_missing = [k for k in missing if "lora_" in k]
if lora_missing:
print(f"[LoRA] Warning: missing LoRA keys (wrong rank/target?): {lora_missing}")
+419
View File
@@ -0,0 +1,419 @@
#!/usr/bin/env python3
"""
LoRA fine-tuning for SelVA / MMAudio generator.
Teaches the model new or partially-known sound classes from custom video+audio pairs.
Only the LoRA adapter weights are trained (~10 MB vs ~4.4 GB for the full model).
Data layout:
data/my_sound/
clip01.mp4 # video files — audio is extracted from the video track
clip02.mp4
prompts.txt # optional: "clip01.mp4: description of the sound"
If prompts.txt is absent, the directory name is used as the prompt for all clips.
Usage:
python train_lora.py \\
--data_dir data/my_sound \\
--output_dir lora_output \\
--variant large_44k \\
--selva_dir /path/to/ComfyUI/models/selva \\
--rank 16 --steps 2000 --lr 1e-4
"""
import argparse
import os
import sys
import math
import random
import json
from pathlib import Path
import torch
import torch.nn.functional as F
import torchaudio
from torchvision.io import read_video
sys.path.insert(0, os.path.dirname(__file__))
from selva_core.model.networks_generator import get_my_mmaudio
from selva_core.model.networks_video_enc import get_my_textsynch
from selva_core.model.utils.features_utils import FeaturesUtils
from selva_core.model.sequence_config import CONFIG_16K, CONFIG_44K
from selva_core.model.flow_matching import FlowMatching
from selva_core.model.lora import apply_lora, get_lora_state_dict
# ---------------------------------------------------------------------------
# Constants (mirror selva_feature_extractor.py)
# ---------------------------------------------------------------------------
_CLIP_SIZE = 384
_SYNC_SIZE = 224
_CLIP_FPS = 8
_SYNC_FPS = 25
_SYNC_MEAN = torch.tensor([0.5, 0.5, 0.5]).view(1, 3, 1, 1)
_SYNC_STD = torch.tensor([0.5, 0.5, 0.5]).view(1, 3, 1, 1)
_VARIANTS = {
"small_16k": ("generator_small_16k_sup_5.pth", "16k", True),
"small_44k": ("generator_small_44k_sup_5.pth", "44k", False),
"medium_44k": ("generator_medium_44k_sup_5.pth", "44k", False),
"large_44k": ("generator_large_44k_sup_5.pth", "44k", False),
}
_VIDEO_EXTS = {".mp4", ".mkv", ".avi", ".mov", ".webm", ".flv"}
# ---------------------------------------------------------------------------
# Data helpers
# ---------------------------------------------------------------------------
def load_prompts(data_dir: Path) -> dict:
"""Load filename → prompt from prompts.txt. Returns empty dict if absent."""
p = data_dir / "prompts.txt"
if not p.exists():
return {}
mapping = {}
for line in p.read_text(encoding="utf-8").splitlines():
line = line.strip()
if not line or line.startswith("#"):
continue
if ":" in line:
fname, prompt = line.split(":", 1)
mapping[fname.strip()] = prompt.strip()
return mapping
def load_clip(path: Path, target_sr: int, duration: float):
"""Load a video file.
Returns:
video: [T, H, W, C] float32 [0, 1]
audio: [L] float32 [-1, 1], resampled and trimmed/padded to duration
source_fps: float
"""
video, audio, info = read_video(str(path), pts_unit="sec", output_format="THWC")
source_fps = float(info.get("video_fps", 30.0))
audio_fps = int(info.get("audio_fps", target_sr))
# Video → float32 [0, 1]
video = video.float() / 255.0 # [T, H, W, C]
# Audio → mono float32 [-1, 1]
target_len = int(duration * target_sr)
if audio.numel() == 0:
audio_out = torch.zeros(target_len)
else:
# audio shape: (channels, samples) — torchvision returns float in [-1, 1]
if audio.dim() == 2:
audio = audio.mean(0) # stereo → mono
elif audio.dim() == 1:
pass
audio = audio.float()
# Safety: clamp to [-1, 1] in case of PCM encoding
if audio.abs().max() > 1.0:
audio = audio / 32768.0
if audio_fps != target_sr:
audio = torchaudio.functional.resample(
audio.unsqueeze(0), audio_fps, target_sr
).squeeze(0)
if audio.shape[0] >= target_len:
audio_out = audio[:target_len]
else:
audio_out = F.pad(audio, (0, target_len - audio.shape[0]))
return video, audio_out, source_fps
def _sample_frames(video, source_fps, target_fps, duration):
T = video.shape[0]
n_out = max(1, int(duration * target_fps))
indices = [min(int(i / target_fps * source_fps), T - 1) for i in range(n_out)]
return video[indices]
def _resize_frames(frames, size):
x = frames.permute(0, 3, 1, 2).float() # [N, C, H, W]
x = F.interpolate(x, size=(size, size), mode="bicubic", align_corners=False)
return x.clamp(0.0, 1.0)
def extract_features(video, audio, source_fps, prompt, duration,
feature_utils, net_video_enc, device, dtype):
"""Extract all conditioning features from a single video+audio clip.
All returned tensors are on CPU, detached — ready to move to device for training.
"""
with torch.no_grad():
# --- Audio latent (VAE encode) ---
# encode_audio is @inference_mode and returns DiagonalGaussianDistribution
audio_b = audio.unsqueeze(0).to(feature_utils.device, dtype) # [1, L]
dist = feature_utils.encode_audio(audio_b)
x1 = dist.mode().clone().cpu() # [1, seq_len, latent_dim] — .clone() exits inference mode
# --- CLIP visual features ---
clip_frames = _sample_frames(video, source_fps, _CLIP_FPS, duration)
clip_frames = _resize_frames(clip_frames, _CLIP_SIZE) # [N, C, 384, 384]
clip_input = clip_frames.unsqueeze(0).to(device, dtype) # [1, N, C, 384, 384]
clip_f = feature_utils.encode_video_with_clip(clip_input).cpu() # [1, N, 1024]
# --- Sync (TextSynchformer) features ---
sync_frames = _sample_frames(video, source_fps, _SYNC_FPS, duration)
sync_frames = _resize_frames(sync_frames, _SYNC_SIZE) # [N, C, 224, 224]
if sync_frames.shape[0] < 16:
pad = 16 - sync_frames.shape[0]
sync_frames = torch.cat(
[sync_frames, sync_frames[-1:].expand(pad, -1, -1, -1)], dim=0)
mean = _SYNC_MEAN.to(sync_frames.device)
std = _SYNC_STD.to(sync_frames.device)
sync_frames = (sync_frames - mean) / std
sync_input = sync_frames.unsqueeze(0).to(device, dtype) # [1, N, C, 224, 224]
text_t5, text_mask = feature_utils.encode_text_t5([prompt])
text_t5, text_mask = net_video_enc.prepend_sup_text_tokens(text_t5, text_mask)
sync_f = net_video_enc.encode_video_with_sync(
sync_input, text_f=text_t5, text_mask=text_mask
).cpu() # [1, T_sync, 768]
# --- CLIP text features ---
text_clip = feature_utils.encode_text_clip([prompt]).cpu() # [1, 77, D]
return x1, clip_f, sync_f, text_clip
# ---------------------------------------------------------------------------
# Main
# ---------------------------------------------------------------------------
def main():
parser = argparse.ArgumentParser(description="LoRA fine-tuning for SelVA generator")
parser.add_argument("--data_dir", required=True, help="Directory with video files and optional prompts.txt")
parser.add_argument("--output_dir", default="lora_output")
parser.add_argument("--variant", default="large_44k", choices=list(_VARIANTS.keys()))
parser.add_argument("--selva_dir", required=True, help="Path to selva model weights (ComfyUI/models/selva)")
parser.add_argument("--rank", type=int, default=16, help="LoRA rank")
parser.add_argument("--alpha", type=float, default=None, help="LoRA alpha (default: rank)")
parser.add_argument("--target", nargs="+", default=["attn.qkv"],
help="Module name suffixes to wrap with LoRA. Also try 'linear1'.")
parser.add_argument("--lr", type=float, default=1e-4)
parser.add_argument("--steps", type=int, default=2000)
parser.add_argument("--warmup_steps",type=int, default=500)
parser.add_argument("--grad_accum", type=int, default=4, help="Gradient accumulation steps")
parser.add_argument("--save_every", type=int, default=500)
parser.add_argument("--precision", default="bf16", choices=["bf16", "fp16", "fp32"])
parser.add_argument("--seed", type=int, default=42)
args = parser.parse_args()
torch.manual_seed(args.seed)
random.seed(args.seed)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
if args.precision == "bf16" and device.type == "cuda" and not torch.cuda.is_bf16_supported():
print("[LoRA] bf16 not supported on this GPU — falling back to fp16")
args.precision = "fp16"
dtype = {"bf16": torch.bfloat16, "fp16": torch.float16, "fp32": torch.float32}[args.precision]
data_dir = Path(args.data_dir)
output_dir = Path(args.output_dir)
selva_dir = Path(args.selva_dir)
output_dir.mkdir(parents=True, exist_ok=True)
gen_filename, mode, has_bigvgan = _VARIANTS[args.variant]
seq_cfg = CONFIG_16K if mode == "16k" else CONFIG_44K
duration = seq_cfg.duration
sample_rate = seq_cfg.sampling_rate
# --- Weight paths ---
def w(name): return str(selva_dir / name)
def wext(name): return str(selva_dir / "ext" / name)
for path, label in [
(w("video_enc_sup_5.pth"), "video_enc"),
(w(gen_filename), "generator"),
(wext("v1-16.pth" if mode == "16k" else "v1-44.pth"), "VAE"),
]:
if not Path(path).exists():
print(f"[LoRA] Missing weight: {path} ({label})")
print("[LoRA] Run ComfyUI with SelvaModelLoader first to auto-download weights.")
sys.exit(1)
synch_path = str(selva_dir / "synchformer_state_dict.pth")
if not Path(synch_path).exists():
# Fallback: check prismaudio dir
alt = selva_dir.parent / "prismaudio" / "synchformer_state_dict.pth"
if alt.exists():
synch_path = str(alt)
else:
print(f"[LoRA] Missing synchformer weights: {synch_path}")
sys.exit(1)
bigvgan_path = wext("best_netG.pt") if has_bigvgan else None
# --- Load models ---
print(f"[LoRA] Loading TextSynch encoder...")
net_video_enc = get_my_textsynch("depth1").to(device, dtype).eval()
net_video_enc.load_weights(
torch.load(w("video_enc_sup_5.pth"), map_location="cpu", weights_only=False)
)
print(f"[LoRA] Loading generator ({args.variant})...")
net_generator = get_my_mmaudio(args.variant).to(device, dtype).eval()
net_generator.load_weights(
torch.load(w(gen_filename), map_location="cpu", weights_only=False)
)
print("[LoRA] Loading FeaturesUtils (need_vae_encoder=True)...")
feature_utils = FeaturesUtils(
tod_vae_ckpt=wext("v1-16.pth" if mode == "16k" else "v1-44.pth"),
synchformer_ckpt=synch_path,
enable_conditions=True,
mode=mode,
bigvgan_vocoder_ckpt=bigvgan_path,
need_vae_encoder=True, # required for audio → latent encoding during training
).to(device, dtype).eval()
# --- Apply LoRA ---
n_lora = apply_lora(
net_generator,
rank=args.rank,
alpha=args.alpha,
target_suffixes=tuple(args.target),
)
print(f"[LoRA] Wrapped {n_lora} linear layers (rank={args.rank}, target={args.target})")
if n_lora == 0:
print("[LoRA] ERROR: no layers were wrapped — check --target names.")
sys.exit(1)
# Freeze everything except LoRA params
for name, p in net_generator.named_parameters():
p.requires_grad_("lora_" in name)
trainable = sum(p.numel() for p in net_generator.parameters() if p.requires_grad)
total = sum(p.numel() for p in net_generator.parameters())
print(f"[LoRA] Trainable: {trainable:,} / {total:,} params "
f"({100 * trainable / total:.2f}%)")
# Update rotary position embeddings for the fixed sequence lengths
net_generator.update_seq_lengths(
latent_seq_len=seq_cfg.latent_seq_len,
clip_seq_len=seq_cfg.clip_seq_len,
sync_seq_len=seq_cfg.sync_seq_len,
)
# --- Dataset ---
video_files = sorted(
p for p in data_dir.iterdir()
if p.suffix.lower() in _VIDEO_EXTS
)
if not video_files:
print(f"[LoRA] No video files found in {data_dir}")
sys.exit(1)
print(f"[LoRA] Found {len(video_files)} video(s) in {data_dir}")
prompt_map = load_prompts(data_dir)
default_prompt = data_dir.name # use directory name as fallback prompt
# Pre-extract features for all clips (cache in RAM)
print("[LoRA] Extracting features from all clips...")
dataset = []
for vf in video_files:
prompt = prompt_map.get(vf.name, default_prompt)
print(f" {vf.name}: '{prompt}'")
try:
video, audio, source_fps = load_clip(vf, sample_rate, duration)
x1, clip_f, sync_f, text_clip = extract_features(
video, audio, source_fps, prompt, duration,
feature_utils, net_video_enc, device, dtype,
)
dataset.append((x1, clip_f, sync_f, text_clip))
except Exception as e:
print(f" [LoRA] Warning: failed to process {vf.name}: {e}")
if not dataset:
print("[LoRA] No clips could be loaded.")
sys.exit(1)
print(f"[LoRA] {len(dataset)} clips ready.")
# --- Optimizer + LR scheduler ---
lora_params = [p for p in net_generator.parameters() if p.requires_grad]
optimizer = torch.optim.AdamW(lora_params, lr=args.lr, weight_decay=1e-2)
def lr_lambda(step):
if step < args.warmup_steps:
return step / max(1, args.warmup_steps)
return 1.0
scheduler = torch.optim.lr_scheduler.LambdaLR(optimizer, lr_lambda)
fm = FlowMatching(min_sigma=0, inference_mode="euler", num_steps=25)
# --- Training loop ---
net_generator.train()
optimizer.zero_grad()
print(f"\n[LoRA] Training: {args.steps} steps, lr={args.lr}, grad_accum={args.grad_accum}")
print(f"[LoRA] Checkpoints every {args.save_every} steps → {output_dir}\n")
total_loss = 0.0
for step in range(1, args.steps + 1):
# Sample a random clip from the dataset
x1_cpu, clip_f_cpu, sync_f_cpu, text_clip_cpu = random.choice(dataset)
x1 = x1_cpu.to(device, dtype)
clip_f = clip_f_cpu.to(device, dtype)
sync_f = sync_f_cpu.to(device, dtype)
text_clip = text_clip_cpu.to(device, dtype)
# Normalize latent in-place (net_generator.normalize is in-place)
net_generator.normalize(x1)
# Flow matching step
t = torch.rand(1, device=device, dtype=dtype) # (1,) — one timestep
x0 = torch.randn_like(x1)
xt = fm.get_conditional_flow(x0, x1, t)
# Forward pass — gradients flow through LoRA A/B only
# forward(latent, clip_f, sync_f, text_f, t) takes raw feature tensors
v_pred = net_generator.forward(xt, clip_f, sync_f, text_clip, t)
loss = fm.loss(v_pred, x0, x1).mean() / args.grad_accum
loss.backward()
total_loss += loss.item() * args.grad_accum
if step % args.grad_accum == 0:
torch.nn.utils.clip_grad_norm_(lora_params, max_norm=1.0)
optimizer.step()
scheduler.step()
optimizer.zero_grad()
if step % 50 == 0:
avg = total_loss / 50
lr_now = scheduler.get_last_lr()[0]
print(f"[LoRA] step {step:5d}/{args.steps} loss={avg:.4f} lr={lr_now:.2e}")
total_loss = 0.0
if step % args.save_every == 0 or step == args.steps:
ckpt = output_dir / f"adapter_step{step:05d}.pt"
torch.save(get_lora_state_dict(net_generator), ckpt)
print(f"[LoRA] Saved {ckpt}")
# Save final adapter with metadata
final = output_dir / "adapter_final.pt"
meta = {
"variant": args.variant,
"rank": args.rank,
"alpha": args.alpha if args.alpha is not None else float(args.rank),
"target": args.target,
"steps": args.steps,
}
torch.save({"state_dict": get_lora_state_dict(net_generator), "meta": meta}, final)
(output_dir / "meta.json").write_text(json.dumps(meta, indent=2))
print(f"\n[LoRA] Training complete. Adapter saved to {final}")
if __name__ == "__main__":
main()