license: mit base_model: moonshotai/Kimi-Linear-48B-A3B-Instruct library_name: transformers pipeline_tag: text-generation tags:

• bf16
• experimental

Kimi-Linear-48B-A3B-Instruct-abliterated-bf16

Public experimental weight-edited checkpoint derived from moonshotai/Kimi-Linear-48B-A3B-Instruct. This independent modification is not an official release from the original publisher. Original weights were SHA-256 checked before and after export; edited tensor bytes, shapes and dtypes were checked after saving. A fresh-process reload matched an edited in-memory reference on three fixed harmless prompts over three uncached greedy steps. These checks establish serialization integrity.

Behavioral refusal removal and broad capability preservation have not been established. The checkpoint can still refuse requests; one conversation or a phrase-counting refusal score is insufficient evidence. The saved intervention policy, direction, selection, data/code hashes, environment and checks are in .pipeline/. Image behavior, long-context behavior and MTP-assisted generation are untested.

Export diagnostics

Held-out harmless first-token KL (100 prompts): 0.075608. Maximum edited/base projection ratio: 0.636%. The 1% mechanical diagnostic passes. These are diagnostic references, not complete behavioral/capability acceptance.

Modification and distribution notice

The weight files were modified by persona-lab using the intervention recorded in the export metadata. Each safetensors file carries a modification notice in its header. This release packaging changes header metadata only: every tensor's dtype, shape, offsets and payload bytes remain the same as the verified edited checkpoint. The original local checkpoint is not rewritten.

Upstream licenses, attribution and usage restrictions continue to apply. See the included license, notice and policy files. No benchmark prompts, raw harmful completions or manual-test programs are included.

Loading

Use persona-lab’s chat_cuda.py and pinned CUDA requirements, including its Kimi compatibility bridge and explicit --trust-remote-code. Example: python chat_cuda.py --model /path/to/downloaded/checkpoint --trust-remote-code --max-tokens 1024. The verified 96 GB GPU configuration uses batch size one and expandable CUDA allocator segments.