⚠ Correction (September 2026): this card described "refusal-free behavior" and 0/40 refusals. That figure came from a small 40-prompt probe and does not reflect measured behavior on a rigorous gate. The abliterated base this adapter sits on measured 40.7% refusal vs official's 36.7% on a 300-prompt gate (temp 1.0); this adapter was never measured at that rigor. Treat its refusal behavior as unknown, not zero.
For a checkpoint with a real measured refusal reduction (~8%), see KidIkaros/abliterated-minicpm5-2b-v2, which supersedes this line entirely (re-ablation + DPO, merged, verified).
A LoRA adapter fine-tuned on an abliterated MiniCPM5-2B, kept for archival purposes.
openbmb/MiniCPM5-2B (abliterated v0)The table below is the original 40-prompt smoke probe (8 questions per category). It is not a refusal-rate measurement — see the correction note above.
| Category | Useful | Refusals |
|---|---|---|
| Refusal Check (harmful queries) | 8/8 | 0/8 |
| Factual QA | 8/8 | 0/8 |
| Coding | 8/8 | 0/8 |
| Math/Science | 8/8 | 0/8 |
| Creative | 8/8 | 0/8 |
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
# Requires the abliterated base model
base = AutoModelForCausalLM.from_pretrained(
"KidIkaros/abliterated-minicpm5-2b-ggml",
torch_dtype="auto",
)
model = PeftModel.from_pretrained(
base,
"KidIkaros/abliterated-minicpm5-2b-lora-finetuned"
)
tokenizer = AutoTokenizer.from_pretrained("openbmb/MiniCPM5-2B")
prompt = "[INST] What is the capital of France? [/INST] "
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
output = model.generate(**inputs, max_new_tokens=128)
print(tokenizer.decode(output[0], skip_special_tokens=True))
This model is a LoRA adapter only. It requires the abliterated base model weights to function:
KidIkaros/abliterated-minicpm5-2b-ggml (GGUF) or local abliterated weightsThis model has been fine-tuned on non-refusal data. It has reduced safety-refusal behavior and is not a safety-aligned product. Users should deploy responsibly.
checkpoint-1000/optimizer.pt 201 MBcheckpoint-1500/optimizer.pt 201 MBcheckpoint-2000/optimizer.pt 201 MBcheckpoint-500/optimizer.pt 201 MBadapter_model.safetensors 101 MBcheckpoint-1000/adapter_model.safetensors 101 MBcheckpoint-1500/adapter_model.safetensors 101 MBcheckpoint-2000/adapter_model.safetensors 101 MBcheckpoint-500/adapter_model.safetensors 101 MBcheckpoint-1000/rng_state.pth 14.6 kBcheckpoint-1500/rng_state.pth 14.6 kBcheckpoint-2000/rng_state.pth 14.6 kBcheckpoint-500/rng_state.pth 14.6 kBcheckpoint-1000/training_args.bin 5.2 kBcheckpoint-1500/training_args.bin 5.2 kBcheckpoint-2000/training_args.bin 5.2 kBcheckpoint-500/training_args.bin 5.2 kBcheckpoint-1000/scheduler.pt 1.5 kBcheckpoint-1500/scheduler.pt 1.5 kBcheckpoint-2000/scheduler.pt 1.5 kBcheckpoint-500/scheduler.pt 1.5 kBcheckpoint-1000/scaler.pt 1.4 kBcheckpoint-1500/scaler.pt 1.4 kBcheckpoint-2000/scaler.pt 1.4 kBcheckpoint-500/scaler.pt 1.4 kB有文件?提交种子。
提交种子openbmb/MiniCPM5-2B