ShadowWork-SHIFT: Plain SFT (exact frozen parent weights)

This is the full, seven-shard Plain SFT model, fine-tuned from CohereLabs/tiny-aya-global. It is the actual base for all three published ShadowWork v2/09 LoRA adapters. Do not substitute the original Cohere checkpoint for this intermediate fine-tune.

Frozen SHA-256 digest (canonical mapping of seven shard hashes): 5a389463f34416e0ec75ace32b07a3090dcb20fcc060d43df9dfa985a37bfd87. See provenance/base_checkpoint_shards_sha256.json for every shard hash.

The model inherits Cohere's CC-BY-NC and Acceptable Use Policy obligations. Intended only for non-commercial research and reproducibility of a synthetic configuration-maintenance tool agent. No real-world operational safety has been established.

Downloads last month
45
Safetensors
Model size
3B params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for i-am-shaurya05/shadowwork-shift-plain-sft

Finetuned
(171)
this model
Adapters
7 models

Collection including i-am-shaurya05/shadowwork-shift-plain-sft