File size: 25,940 Bytes
e527a65
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
import argparse
import json
import inspect  # Added for Transformers version compatibility
import math
import time
from pathlib import Path
from typing import Any, Dict, Optional, Tuple, List

import torch
import yaml
from datasets import load_dataset, DatasetDict
from huggingface_hub import snapshot_download
from transformers import (
    AutoModelForCausalLM,
    AutoTokenizer,
    PreTrainedTokenizerFast,
    TrainingArguments,
    Trainer,
    TrainerCallback,
    default_data_collator,
    set_seed,
)
from transformers.trainer_utils import get_last_checkpoint
from peft import (
    LoraConfig,
    get_peft_model,
    prepare_model_for_kbit_training,
    PeftModel,
)

try:
    from transformers import BitsAndBytesConfig
except ImportError:  # older transformers
    BitsAndBytesConfig = None


# --------------------------
# Helpers
# --------------------------

def _dtype_from_str(s: str) -> torch.dtype:
    s = (s or "").lower()
    if s in ("float16", "fp16"):
        return torch.float16
    if s in ("bfloat16", "bf16"):
        return torch.bfloat16
    if s in ("float32", "fp32"):
        return torch.float32
    raise ValueError(f"Unknown torch_dtype: {s}")

def _now_iso() -> str:
    return time.strftime("%Y-%m-%dT%H:%M:%S", time.localtime())

def _safe_exp(x: float) -> float:
    x = min(float(x), 50.0)
    return float(math.exp(x))

def _ensure_dir(p: Path) -> Path:
    p.mkdir(parents=True, exist_ok=True)
    return p

def _looks_like_model_dir(p: Path) -> bool:
    if not p.exists() or not p.is_dir():
        return False
    if (p / "config.json").exists():
        return True
    if any(p.glob("*.safetensors")) or any(p.glob("pytorch_model*.bin")):
        return True
    return False

def _detect_text_field(example: Dict[str, Any]) -> Optional[str]:
    for k, v in example.items():
        if isinstance(v, str) and v.strip():
            return k
    return None

def _load_tokenizer(base_dir: Path, use_fast: bool, trust_remote_code: bool):
    try:
        return AutoTokenizer.from_pretrained(
            str(base_dir),
            use_fast=use_fast,
            trust_remote_code=trust_remote_code,
        )
    except ValueError as e:
        if "TokenizersBackend" not in str(e):
            raise
        tok_file = base_dir / "tokenizer.json"
        tok_cfg_path = base_dir / "tokenizer_config.json"
        if not tok_file.exists():
            raise

        tok_kwargs: Dict[str, Any] = {}
        if tok_cfg_path.exists():
            with tok_cfg_path.open("r", encoding="utf-8") as f:
                tok_cfg = json.load(f)
            for key in ("bos_token", "eos_token", "pad_token", "unk_token", "model_max_length"):
                if tok_cfg.get(key) is not None:
                    tok_kwargs[key] = tok_cfg[key]
            extra = tok_cfg.get("additional_special_tokens") or tok_cfg.get("extra_special_tokens")
            if extra:
                tok_kwargs["additional_special_tokens"] = extra

        return PreTrainedTokenizerFast(tokenizer_file=str(tok_file), **tok_kwargs)

def _infer_target_modules(model) -> List[str]:
    names = set()
    for n, _ in model.named_modules():
        names.add(n.split(".")[-1])

    for group in [
        ["q_proj", "k_proj", "v_proj", "o_proj"],
        ["Wqkv", "out_proj"],
        ["query_key_value", "dense"],
        ["c_attn", "c_proj"],
    ]:
        if all(x in names for x in group):
            return group

    fallback = [x for x in ["q_proj", "k_proj", "v_proj", "o_proj", "c_attn", "c_proj", "out_proj", "dense"] if x in names]
    if fallback:
        return fallback

    raise ValueError("Could not auto-infer target_modules. Set peft.target_modules explicitly.")

def _choose_attn_impl(cfg: Dict[str, Any]) -> Optional[str]:
    return cfg.get("model", {}).get("attn_implementation", None)


# --------------------------
# JSONL Logger Callback
# --------------------------

class JsonlLoggerCallback(TrainerCallback):
    def __init__(self, run_dir: Path):
        self.run_dir = run_dir
        self.train_log_path = _ensure_dir(run_dir / "logs") / "train.jsonl"
        self.eval_log_path = _ensure_dir(run_dir / "logs") / "eval.jsonl"
        self.start_time = None

    def _eta(self, global_step: int, max_steps: int) -> Optional[str]:
        if self.start_time is None or global_step <= 0 or max_steps <= 0:
            return None
        elapsed = time.time() - self.start_time
        sec_per_step = elapsed / global_step
        remaining = max(0, max_steps - global_step) * sec_per_step
        h = int(remaining // 3600)
        m = int((remaining % 3600) // 60)
        s = int(remaining % 60)
        return f"{h:02d}:{m:02d}:{s:02d}"

    def on_train_begin(self, args, state, control, **kwargs):
        self.start_time = time.time()

    def on_log(self, args, state, control, logs=None, **kwargs):
        if not logs:
            return

        max_steps = int(state.max_steps) if getattr(state, "max_steps", None) else 0
        progress_pct = (100.0 * state.global_step / max_steps) if max_steps > 0 else None
        epoch_pct = None
        if state.epoch is not None and args.num_train_epochs and args.num_train_epochs > 0:
            epoch_pct = 100.0 * (float(state.epoch) / float(args.num_train_epochs))

        payload = {
            "ts": _now_iso(),
            "event": "train_log",
            "step": int(state.global_step),
            "epoch": round(float(state.epoch), 4) if state.epoch is not None else None,
            "progress_pct": round(progress_pct, 2) if progress_pct is not None else None,
            "epoch_pct": round(epoch_pct, 2) if epoch_pct is not None else None,
            "eta": self._eta(int(state.global_step), max_steps),
            "max_grad_norm": getattr(args, "max_grad_norm", None),
            **logs,
        }

        with self.train_log_path.open("a", encoding="utf-8") as f:
            f.write(json.dumps(payload, ensure_ascii=False) + "\n")

    def on_evaluate(self, args, state, control, metrics=None, **kwargs):
        if not metrics:
            return
        eval_loss = metrics.get("eval_loss", None)
        ppl = _safe_exp(eval_loss) if eval_loss is not None else None

        payload = {
            "ts": _now_iso(),
            "event": "eval",
            "step": int(state.global_step),
            "epoch": float(state.epoch) if state.epoch is not None else None,
            **metrics,
            "perplexity": ppl,
        }
        with self.eval_log_path.open("a", encoding="utf-8") as f:
            f.write(json.dumps(payload, ensure_ascii=False) + "\n")


# --------------------------
# Data Pipeline (EOS + Packing)
# --------------------------

def build_datasets(cfg: Dict[str, Any], tokenizer) -> Tuple[Any, Any]:
    data_cfg = cfg["data"]
    train_path = data_cfg["train_jsonl"]
    eval_path = data_cfg.get("eval_jsonl", None)
    split_ratio = float(data_cfg.get("eval_split_ratio", 0.0))
    text_field = data_cfg.get("text_field", "text")
    block_size = int(data_cfg.get("block_size", 2048))
    shuffle = bool(data_cfg.get("shuffle", True))
    num_proc = int(data_cfg.get("num_proc", 4))

    pack_mode = str(data_cfg.get("pack_mode", "drop")).lower().strip()
    if pack_mode not in ("drop", "pad"):
        raise ValueError(f"data.pack_mode must be 'drop' or 'pad', got: {pack_mode}")

    eos_id = tokenizer.eos_token_id
    if eos_id is None:
        raise ValueError("Tokenizer has no eos_token_id; CPT packing needs an EOS delimiter.")

    if tokenizer.pad_token_id is None:
        # safe default for many causal LMs
        tokenizer.pad_token = tokenizer.eos_token
    pad_id = tokenizer.pad_token_id

    ds = load_dataset("json", data_files={"train": train_path})

    if eval_path:
        ds_eval = load_dataset("json", data_files={"eval": eval_path})
        dsd = DatasetDict({"train": ds["train"], "eval": ds_eval["eval"]})
    else:
        if 0.0 < split_ratio < 1.0:
            split = ds["train"].train_test_split(test_size=split_ratio, seed=int(cfg["run"].get("seed", 42)))
            dsd = DatasetDict({"train": split["train"], "eval": split["test"]})
        else:
            dsd = DatasetDict({"train": ds["train"], "eval": None})

    if text_field not in dsd["train"].column_names:
        auto_field = _detect_text_field(dsd["train"][0])
        if not auto_field:
            raise ValueError(f"Could not find text field. Columns: {dsd['train'].column_names}")
        text_field = auto_field

    def tokenize_fn(examples):
        out = tokenizer(
            examples[text_field],
            add_special_tokens=False,
            truncation=False,
            padding=False,
        )
        if "token_type_ids" in out:
            del out["token_type_ids"]
        # Add EOS between docs
        out["input_ids"] = [ids + [eos_id] for ids in out["input_ids"]]
        out["attention_mask"] = [m + [1] for m in out["attention_mask"]]
        return out

    tokenized_train = dsd["train"].map(
        tokenize_fn,
        batched=True,
        num_proc=num_proc,
        remove_columns=dsd["train"].column_names,
        desc="Tokenizing train",
    )

    tokenized_eval = None
    if dsd["eval"] is not None:
        tokenized_eval = dsd["eval"].map(
            tokenize_fn,
            batched=True,
            num_proc=num_proc,
            remove_columns=dsd["eval"].column_names,
            desc="Tokenizing eval",
        )

    def group_texts(examples):
        concatenated = {k: sum(examples[k], []) for k in examples.keys()}
        total_length = len(concatenated["input_ids"])

        if total_length == 0:
            return {"input_ids": [], "attention_mask": [], "labels": []}

        full_len = (total_length // block_size) * block_size
        blocks_input, blocks_attn, blocks_labels = [], [], []

        # full blocks
        for i in range(0, full_len, block_size):
            chunk = concatenated["input_ids"][i:i + block_size]
            attn = concatenated["attention_mask"][i:i + block_size]
            blocks_input.append(chunk)
            blocks_attn.append(attn)
            blocks_labels.append(chunk.copy())

        # remainder
        remainder = total_length - full_len
        if remainder > 0 and pack_mode == "pad":
            chunk = concatenated["input_ids"][full_len:full_len + remainder]
            attn = concatenated["attention_mask"][full_len:full_len + remainder]

            pad_len = block_size - remainder
            chunk_padded = chunk + [pad_id] * pad_len
            attn_padded = attn + [0] * pad_len

            labels = chunk_padded.copy()
            labels[-pad_len:] = [-100] * pad_len  # loss mask

            blocks_input.append(chunk_padded)
            blocks_attn.append(attn_padded)
            blocks_labels.append(labels)

        return {
            "input_ids": blocks_input,
            "attention_mask": blocks_attn,
            "labels": blocks_labels,
        }

    tokenized_train = tokenized_train.map(
        group_texts,
        batched=True,
        num_proc=num_proc,
        desc=f"Packing train blocks (mode={pack_mode})",
    )
    if tokenized_eval is not None:
        tokenized_eval = tokenized_eval.map(
            group_texts,
            batched=True,
            num_proc=num_proc,
            desc=f"Packing eval blocks (mode={pack_mode})",
        )

    if len(tokenized_train) == 0:
        raise ValueError(
            "Train dataset is empty after packing. "
            "Either increase data, reduce block_size, or set data.pack_mode='pad'."
        )

    if shuffle:
        tokenized_train = tokenized_train.shuffle(seed=int(cfg["run"].get("seed", 42)))

    return tokenized_train, tokenized_eval


# --------------------------
# Model Loading + PEFT
# --------------------------

def _select_model_loader(base_dir: Path):
    cfg_path = base_dir / "config.json"
    if not cfg_path.exists():
        return {"kind": "causal", "arch": None}
    with cfg_path.open("r", encoding="utf-8") as f:
        cfg = json.load(f)
    arch = cfg.get("architectures") or []
    arch_name = arch[0] if arch else None
    if any("ForConditionalGeneration" in a for a in arch):
        return {"kind": "conditional", "arch": arch_name}
    return {"kind": "causal", "arch": arch_name}

def _resolve_model_class(arch_name: str):
    import transformers
    cls = getattr(transformers, arch_name, None)
    if cls is None:
        raise ValueError(f"Model class '{arch_name}' is not available in installed transformers.")
    return cls


def load_base_model_and_tokenizer(cfg: Dict[str, Any], base_dir: Path):
    model_cfg = cfg["model"]
    trust_remote_code = bool(model_cfg.get("trust_remote_code", True))
    use_fast = bool(model_cfg.get("tokenizer_use_fast", True))
    device_map = model_cfg.get("device_map", "auto")

    tokenizer = _load_tokenizer(base_dir, use_fast=use_fast, trust_remote_code=trust_remote_code)
    if tokenizer.pad_token is None:
        tokenizer.pad_token = tokenizer.eos_token

    torch_dtype = _dtype_from_str(model_cfg.get("torch_dtype", "bfloat16"))
    use_4bit = bool(model_cfg.get("use_4bit", False))

    quant_cfg = None
    if use_4bit:
        if BitsAndBytesConfig is None:
            raise ImportError("BitsAndBytesConfig is not available in this transformers version.")
        quant_cfg = BitsAndBytesConfig(
            load_in_4bit=True,
            bnb_4bit_quant_type=str(model_cfg.get("bnb_4bit_quant_type", "nf4")),
            bnb_4bit_use_double_quant=bool(model_cfg.get("bnb_4bit_use_double_quant", True)),
            bnb_4bit_compute_dtype=_dtype_from_str(model_cfg.get("bnb_4bit_compute_dtype", "bfloat16")),
        )

    attn_impl = _choose_attn_impl(cfg)
    model_meta = _select_model_loader(base_dir)

    try:
        if model_meta["kind"] == "conditional":
            model_cls = _resolve_model_class(model_meta["arch"]) if model_meta["arch"] else None
            if model_cls is None:
                raise ValueError("Conditional model architecture not specified in config.json.")
            model = model_cls.from_pretrained(
                str(base_dir),
                device_map=device_map,
                trust_remote_code=trust_remote_code,
                low_cpu_mem_usage=True,
                torch_dtype=(torch_dtype if not use_4bit else None),
                quantization_config=quant_cfg,
                attn_implementation=attn_impl,
            )
        else:
            model = AutoModelForCausalLM.from_pretrained(
                str(base_dir),
                device_map=device_map,
                trust_remote_code=trust_remote_code,
                low_cpu_mem_usage=True,
                torch_dtype=(torch_dtype if not use_4bit else None),
                quantization_config=quant_cfg,
                attn_implementation=attn_impl,
            )
    except Exception as e:
        if attn_impl is not None:
            print(f"[warn] attn_implementation='{attn_impl}' failed: {e}")
            print("[warn] Falling back to default attention implementation.")
        if model_meta["kind"] == "conditional":
            model_cls = _resolve_model_class(model_meta["arch"]) if model_meta["arch"] else None
            if model_cls is None:
                raise ValueError("Conditional model architecture not specified in config.json.")
            model = model_cls.from_pretrained(
                str(base_dir),
                device_map=device_map,
                trust_remote_code=trust_remote_code,
                low_cpu_mem_usage=True,
                torch_dtype=(torch_dtype if not use_4bit else None),
                quantization_config=quant_cfg,
            )
        else:
            model = AutoModelForCausalLM.from_pretrained(
                str(base_dir),
                device_map=device_map,
                trust_remote_code=trust_remote_code,
                low_cpu_mem_usage=True,
                torch_dtype=(torch_dtype if not use_4bit else None),
                quantization_config=quant_cfg,
            )

    return model, tokenizer


def apply_peft(cfg: Dict[str, Any], model):
    peft_cfg = cfg["peft"]
    model_cfg = cfg["model"]
    tr_cfg = cfg["train"]

    if not bool(peft_cfg.get("enabled", True)):
        return model, None

    use_4bit = bool(model_cfg.get("use_4bit", False))
    gradient_checkpointing = bool(tr_cfg.get("gradient_checkpointing", True))

    if gradient_checkpointing and hasattr(model, "gradient_checkpointing_enable"):
        model.gradient_checkpointing_enable()
        if hasattr(model, "config"):
            model.config.use_cache = False

    if use_4bit:
        model = prepare_model_for_kbit_training(
            model,
            use_gradient_checkpointing=gradient_checkpointing,
        )

    target_modules = peft_cfg.get("target_modules", "auto")
    if target_modules == "auto":
        target_modules = _infer_target_modules(model)

    lora_config = LoraConfig(
        r=int(peft_cfg.get("r", 16)),
        lora_alpha=int(peft_cfg.get("lora_alpha", 32)),
        lora_dropout=float(peft_cfg.get("lora_dropout", 0.05)),
        bias=str(peft_cfg.get("bias", "none")),
        task_type="CAUSAL_LM",
        target_modules=target_modules,
    )
    model = get_peft_model(model, lora_config)
    return model, lora_config


# --------------------------
# Merge Logic
# --------------------------

def merge_adapter(cfg: Dict[str, Any], base_dir: Path, adapter_dir: Path, final_dir: Path):
    print(f"--- Merge: {adapter_dir} + {base_dir} -> {final_dir} ---")

    model_cfg = cfg["model"]
    merge_cfg = cfg.get("merge", {})
    trust_remote_code = bool(model_cfg.get("trust_remote_code", True))
    use_fast = bool(model_cfg.get("tokenizer_use_fast", True))

    merged_dtype = _dtype_from_str(merge_cfg.get("merged_dtype", "float16"))
    max_shard_size = str(merge_cfg.get("max_shard_size", "2GB"))

    model_meta = _select_model_loader(base_dir)
    if model_meta["kind"] == "conditional":
        base_cls = _resolve_model_class(model_meta["arch"]) if model_meta["arch"] else None
        if base_cls is None:
            raise ValueError("Conditional model architecture not specified in config.json.")
        base = base_cls.from_pretrained(
            str(base_dir),
            torch_dtype=merged_dtype,
            device_map="cpu",
            low_cpu_mem_usage=True,
            trust_remote_code=trust_remote_code,
        )
    else:
        base = AutoModelForCausalLM.from_pretrained(
            str(base_dir),
            torch_dtype=merged_dtype,
            device_map="cpu",
            low_cpu_mem_usage=True,
            trust_remote_code=trust_remote_code,
        )

    merged = PeftModel.from_pretrained(base, str(adapter_dir))
    merged = merged.merge_and_unload()

    _ensure_dir(final_dir)
    # Fix for transformers weight conversion bug with quantized models
    # Clear weight conversions to avoid NotImplementedError in reverse_transform
    if hasattr(merged, '_weight_conversions'):
        merged._weight_conversions = []
    merged.save_pretrained(
        str(final_dir), 
        safe_serialization=True, 
        max_shard_size=max_shard_size
    )

    tok = _load_tokenizer(base_dir, use_fast=use_fast, trust_remote_code=trust_remote_code)
    if tok.pad_token is None:
        tok.pad_token = tok.eos_token
    tok.save_pretrained(str(final_dir))

    print("--- Merge complete ---")


# --------------------------
# Main
# --------------------------

def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--config", required=True, help="Path to YAML config")
    ap.add_argument("--merge-only", action="store_true", help="Skip training, just merge adapter")
    args = ap.parse_args()

    with open(args.config, "r", encoding="utf-8") as f:
        cfg = yaml.safe_load(f)

    run_dir = _ensure_dir(Path(cfg["run"]["run_dir"]))
    _ensure_dir(run_dir / "logs")

    with (run_dir / "config_resolved.yaml").open("w", encoding="utf-8") as f:
        yaml.safe_dump(cfg, f, sort_keys=False)

    model_cfg = cfg["model"]
    repo_id = str(model_cfg["repo_id"]).strip()
    repo_path = Path(repo_id)

    # ✅ Local model path -> load directly; no download
    if repo_path.exists() and repo_path.is_dir():
        base_dir = repo_path
        if not _looks_like_model_dir(base_dir):
            raise ValueError(f"model.repo_id points to a directory, but it doesn't look like a HF model dir: {base_dir}")
    else:
        # HF repo_id -> download into run_dir/base_local_dir
        base_dir = _ensure_dir(run_dir / model_cfg.get("base_local_dir", "base_model"))
        if not _looks_like_model_dir(base_dir):
            print(f"Base model not found at {base_dir}, downloading from {repo_id} ...")
            snapshot_download(
                repo_id=repo_id,
                revision=model_cfg.get("revision", None),
                local_dir=str(base_dir),
                local_dir_use_symlinks=False,
            )

    ckpt_dir = _ensure_dir(run_dir / "checkpoints")
    best_adapter_dir = _ensure_dir(run_dir / "best_adapter")

    merge_cfg = cfg.get("merge", {}) or {}
    if merge_cfg.get("output_dir"):
        od = Path(str(merge_cfg["output_dir"]))
        final_dir = od if od.is_absolute() else (run_dir / od)
    else:
        final_dir = run_dir / "final_model"

    # Merge-only
    if args.merge_only:
        if not _looks_like_model_dir(best_adapter_dir):
            raise FileNotFoundError(f"Adapter not found at {best_adapter_dir}")
        merge_adapter(cfg, base_dir, best_adapter_dir, final_dir)
        return

    # Training
    set_seed(int(cfg["run"].get("seed", 42)))

    model, tokenizer = load_base_model_and_tokenizer(cfg, base_dir)
    model, _ = apply_peft(cfg, model)

    train_ds, eval_ds = build_datasets(cfg, tokenizer)

    tr_cfg = cfg["train"]

    dtype = _dtype_from_str(model_cfg.get("torch_dtype", "bfloat16"))
    use_fp16 = (dtype == torch.float16)
    use_bf16 = (dtype == torch.bfloat16)

    max_steps = int(tr_cfg.get("max_steps", 0))
    num_train_epochs = float(tr_cfg.get("num_train_epochs", 1))

    # --- Dynamic evaluation strategy parameter handling ---
    ta_params = inspect.signature(TrainingArguments.__init__).parameters
    eval_key = "eval_strategy" if "eval_strategy" in ta_params else "evaluation_strategy"

    desired_ta_kwargs = dict(
        output_dir=str(ckpt_dir),
        max_steps=max_steps if max_steps > 0 else -1,
        num_train_epochs=num_train_epochs,

        per_device_train_batch_size=int(tr_cfg.get("per_device_train_batch_size", 1)),
        per_device_eval_batch_size=int(tr_cfg.get("per_device_eval_batch_size", tr_cfg.get("per_device_train_batch_size", 1))),
        gradient_accumulation_steps=int(tr_cfg.get("gradient_accumulation_steps", 1)),

        learning_rate=float(tr_cfg.get("learning_rate", 2e-5)),
        weight_decay=float(tr_cfg.get("weight_decay", 0.0)),
        warmup_ratio=float(tr_cfg.get("warmup_ratio", 0.0)),
        lr_scheduler_type=str(tr_cfg.get("lr_scheduler_type", "cosine")),

        optim=str(tr_cfg.get("optim", "paged_adamw_8bit" if bool(model_cfg.get("use_4bit", False)) else "adamw_torch")),
        max_grad_norm=float(tr_cfg.get("max_grad_norm", 1.0)),

        logging_steps=int(tr_cfg.get("logging_steps", 10)),

        save_strategy=str(tr_cfg.get("save_strategy", "steps")),
        save_steps=int(tr_cfg.get("save_steps", 200)),
        save_total_limit=int(tr_cfg.get("save_total_limit", 3)),

        eval_steps=int(tr_cfg.get("eval_steps", 200)),

        load_best_model_at_end=bool(tr_cfg.get("load_best_model_at_end", True)) if eval_ds is not None else False,
        metric_for_best_model="eval_loss",
        greater_is_better=False,

        fp16=use_fp16,
        bf16=use_bf16,

        report_to=[],
        remove_unused_columns=False,
        save_safetensors=True,
        overwrite_output_dir=False,
    )

    # Set the correct argument name for this transformers version
    desired_ta_kwargs[eval_key] = str(tr_cfg.get("evaluation_strategy", "steps" if eval_ds is not None else "no"))
    ta_kwargs = {k: v for k, v in desired_ta_kwargs.items() if k in ta_params}

    training_args = TrainingArguments(**ta_kwargs)

    trainer_params = inspect.signature(Trainer.__init__).parameters
    desired_trainer_kwargs = dict(
        model=model,
        args=training_args,
        train_dataset=train_ds,
        eval_dataset=eval_ds,
        tokenizer=tokenizer,
        processing_class=tokenizer,
        data_collator=default_data_collator,
        callbacks=[JsonlLoggerCallback(run_dir)],
    )
    trainer_kwargs = {k: v for k, v in desired_trainer_kwargs.items() if k in trainer_params}
    trainer = Trainer(**trainer_kwargs)

    # Resume
    resume_from = tr_cfg.get("resume_from_checkpoint", None)
    if resume_from == "auto":
        last = get_last_checkpoint(str(ckpt_dir))
        resume_from = last if last else None
        if resume_from:
            print(f"Resuming from {resume_from}")

    print("Starting training...")
    trainer.train(resume_from_checkpoint=resume_from)

    trainer.save_model(str(best_adapter_dir))
    print(f"Saved best adapter -> {best_adapter_dir}")

    if eval_ds is not None:
        metrics = trainer.evaluate()
        eval_loss = metrics.get("eval_loss", None)
        metrics["perplexity"] = _safe_exp(eval_loss) if eval_loss is not None else None
        with (run_dir / "eval_final.json").open("w", encoding="utf-8") as f:
            json.dump(metrics, f, indent=2)
        print(f"Final eval_loss={eval_loss}, ppl={metrics['perplexity']}")

    if bool(cfg.get("merge", {}).get("enabled", False)):
        del trainer, model
        torch.cuda.empty_cache()
        merge_adapter(cfg, base_dir, best_adapter_dir, final_dir)
    else:
        print("Merge disabled. Run with --merge-only later if needed.")


if __name__ == "__main__":
    main()