LFM2.5-24B-A2B
LiquidAI/LFM2.5-24B-A2B
24B · 2B activeLFM5 Heretic artifacts1 measured conversions1.5K downloads
Best version49.6Pareto
OS-Software/LFM2.5-24B-A2B-Uncensored-Heretic-GGUFby OS-Software · GGUF, GPTQ, SAFETENSORS
- KL divergence
- 0.017
- Refusals
- 99/100
- Refusal rate
- 99%
- Downloads
- 590
Evidence from the model card
| File | Refusals (original → this) | KL divergence | MMLU vs original | → | QAD Q40 | 99/100 → 0/100 | 0.0174 | −0.62 (all 14,042, p = 0.004) |
Metrics published by the author. Why this ranks first.
All measured versions
Higher score is better. One row per ablation attempt; artifacts of the same attempt are grouped under it.
| # | Author | Params | KL | Refusals | Score | Evidence | Downloads |
|---|---|---|---|---|---|---|---|
| 1 | OS-Software | 24B · 2B active | 0.017 | 99/10099% | 49.6 | BestParetoGGUFGPTQSAFETENSORS | 590 |
All artifacts (5)
| Repository | Type | Formats | Metrics | Downloads | Updated |
|---|---|---|---|---|---|
| LFM2.5-24B-A2B-Abliterated-GGUF alexwkleung | heretic | GGUFQ4_K_M | none | 847 | 2026-10-15 |
| LFM2.5-24B-A2B-Uncensored-Heretic-GGUF OS-Software | quant | GGUFQ4_K_M | published (weak match)KL 0.017 | 403 | 2026-10-16 |
| LFM2.5-24B-A2B-Uncensored-Heretic OS-Software | finetune | SAFETENSORS | published (weak match)KL 0.017 | 187 | 2026-10-16 |
| LFM2.5-24B-A2B-Uncensored-Heretic-GGUF mradermacher | quant | GGUFIQ4_XS | none | 68 | 2026-10-17 |
| LFM2.5-24B-A2B-Uncensored-Heretic-i1-GGUF mradermacher | quant | GGUFIQ1_M | none | 36 | 2026-10-17 |
3 artifacts inherit metrics from their parent conversion or publish none.