fix: remove non-existent add_entry call in _commit_changes

fix: _metrics -> metrics in _commit_changes

fix: batch=2 accum=1 for T4 VRAM

fix: hash only model.safetensors+config.json for upload skip detection (was hashing entire tmpdir incl checkpoints/charts)

maintenance: atomic checkpoints, error logging, dead code removal, config alignment, KV cache cap, robustness fixes

fix: compiled model truthiness in evolution bug_diagnosis

fix: compiled model truthiness check (is not None, not __len__)

fix: batch=20 accum=2 to fit T4 VRAM with torch.compile

fix: torch.compile force_parameter_static_shapes=False for bitnet weight shapes

fix: torch.compile backend=eager (inductor incompatible with buffer mutations)

fix: pass num_heads to MLAConfig; always update flat_grad on shape mismatch

feat: torch.compile, batch=40 accum=1

fix: skip empty HF commits via content hash check

fix: use pre-clip grad norm; faster cos_sim EMA (0.3), skip stale flat_grad on shape mismatch

fix: recursive nested dataclass update in config loading

fix: HybridAttention missing rope_theta/max_seq_len attributes

bug fixes & improvements: chart string-safe, HF/kaggle_train split, autonomous module, curriculum router guards

fix: include kaggle_train.py in dataset sync

fix: missing _analyze_logs method

fix: guard curriculum_router missing methods

fix: HF serialization with _clean_for_json

fix: total_tokens resume offset

full self-evolution: trainer integration, real changes, auto episodes

update deeplm/ code, remove data/

Initial upload: source + config + tokenizer + charts + metrics

update deeplm/ source code (full: model, training, quantization, self_evolution, data, inference)

Delete deeplm/scripts/train_bitnet.py with huggingface_hub

Delete deeplm/scripts/save_model.py with huggingface_hub

Delete deeplm/deeplm/training/trainer.py with huggingface_hub

Delete deeplm/deeplm/training/logger.py with huggingface_hub

Delete deeplm/deeplm/training/data_pipeline.py with huggingface_hub

Delete deeplm/deeplm/training/curriculum_router.py with huggingface_hub

Delete deeplm/deeplm/training/control/training_control.py with huggingface_hub

Delete deeplm/deeplm/training/control/__init__.py with huggingface_hub

Delete deeplm/deeplm/training/auto_tuner.py with huggingface_hub

Delete deeplm/deeplm/training/__init__.py with huggingface_hub

Delete deeplm/deeplm/self_evolution/framework.py with huggingface_hub

Delete deeplm/deeplm/self_evolution/__init__.py with huggingface_hub

Delete deeplm/deeplm/quantization/gguf_export.py with huggingface_hub

Delete deeplm/deeplm/quantization/bitnet_quantize.py with huggingface_hub

Delete deeplm/deeplm/quantization/__init__.py with huggingface_hub

Delete deeplm/deeplm/model/transformer_block.py with huggingface_hub

Delete deeplm/deeplm/model/mtp.py with huggingface_hub

Delete deeplm/deeplm/model/moe.py with huggingface_hub

Delete deeplm/deeplm/model/mla.py with huggingface_hub

Delete deeplm/deeplm/model/hyper_connections.py with huggingface_hub

Delete deeplm/deeplm/model/hybrid_attention.py with huggingface_hub

Delete deeplm/deeplm/model/deeplm.py with huggingface_hub

Delete deeplm/deeplm/model/__init__.py with huggingface_hub

Delete deeplm/deeplm/inference/generate.py with huggingface_hub