perf: use torch.compile max-autotune mode

perf: remove CPU-GPU sync bottleneck in SharedMoE routing loop

Restore full README with training history, swarm table, and specialist status

Add instruction-tuned model in safetensors format

Delete model.safetensors

Add model card

Add generate.py from base model

Add modeling_eve.py from base model

Add configuration_eve.py from base model

Add generation_config.json from base model

Add config.json from base model

Add instruction-tuned weights (3 epochs on alpaca-cleaned)

Fix: move super().__init__() before attribute assignments to prevent PretrainedConfig clobbering MoE top_k

Update README.md

Update model card with Eve-2 logo

Update README.md

Add Eve-2 swarm logo

Update README.md

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Delete make_nb.py

Upload folder using huggingface_hub

Delete push_to_hub.py

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Create README.md

Eve-2-MoE-IT-272M: heavy IT patch (open-perfectblend, LoRA r=128, merged)

initial commit