cd /news/artificial-intelligence/atomicchat-strips-lfm2-5-refusals-fr… · home › topics › artificial-intelligence › article
[ARTICLE · art-170662] src=alphasignal.ai ↗ pub= topic=artificial-intelligence verified=true sentiment=· neutral

AtomicChat Strips LFM2.5 Refusals From 96% to 1% by Editing Later Blocks

AtomicChat released refusal-ablated GGUF builds of Liquid AI's LFM2.5-24B-A2B mixture-of-experts model on Hugging Face, cutting English refusal rates from 96.3% to 1.2% while MMLU rose from 64.40% to 65.55%. AtomicChat edited only blocks 14 through 39 of the model's 40 layers, reporting that touching all 40 layers cost 4 MMLU points, and shipped six quants ranging from 12.1 GB to 25.7 GB plus a 24 MB LoRA adapter for existing quants. The repository has passed 800,000 downloads, a counter that measures file downloads rather than unique users, and AtomicChat says the method extends Arditi et al.'s refusal-direction work to sparse MoE architectures.

read2 min views3 publishedNov 16, 2026
AtomicChat Strips LFM2.5 Refusals From 96% to 1% by Editing Later Blocks
Image: Alphasignal (auto-discovered)

AtomicChat shipped refusal-ablated GGUF builds of Liquid AI's LFM2.5-24B-A2B MoE, cutting English refusals from 96.3% to 1.2% while gaining MMLU points.

  • AtomicChat released refusal-ablated GGUF builds of Liquid AI's LFM2.5-24B-A2B MoE model

  • English refusals dropped from 96.3% to 1.2% while MMLU rose from 64.40% to 65.55%

  • Only blocks 14-39 edited; touching all 40 layers cost 4 MMLU points

  • Six quants available (12.1 GB to 25.7 GB) plus a 24 MB LoRA adapter for existing quants

  • Base model runs at 184 tokens/s on M5 Max with 131K context and 2B active params

  • Method extends Arditi et al. refusal direction work to sparse MoE architectures AtomicChat has published six quantized GGUF builds and a compact LoRA adapter that suppress refusals in Liquid AI’s LFM2.5-24B-A2B, a sparse mixture-of-experts model designed for local agents. The Hugging Face repository has passed 800,000 downloads, though that counter measures file downloads rather than unique users. The release’s main technical finding is that refusal behavior appears concentrated in later MoE expert weights, beyond the mixer outputs targeted by many existing editing tools.

A 24B model that activates a fraction of its weights #

The LFM base model stores 24 billion parameters while activating roughly 2 billion per token. Its 40-layer architecture combines 30 double-gated short-convolution blocks with 10 grouped-query-attention blocks. Within each MoE block, a router sends each token to four of 64 experts, reducing computation to about 2.45 billion parameters per forward pass.

Liquid AI reports performance up to 1.7 times faster than comparable on-device MoEs, including 184 tokens per second on an Apple M5 Max and up to 21,000 output tokens per second on one NVIDIA B300. Those figures come from different hardware and workload configurations, so actual throughput will depend on quantization, batch size, context length, memory bandwidth, and runtime settings.

The edit starts at block 14 #

AtomicChat reports a sharp reduction in refusals after editing blocks 14 through 39 and leaving the earlier network unchanged. Its held-out evaluation produced the following results:

Measure Base model Edited model
English refusal rate 96.3% 1.2%
Refusals with thinking enabled 30 of 30 0 of 30
MMLU score 64.40 65.55

A single MMLU result supports retention of broad knowledge performance, although it cannot establish parity across calibration, long-context tasks, languages, tool use, or specialized domains. The modest score increase may also fall within evaluation variance. Independent reproduction would strengthen the publisher-reported results.

This story is for Pro members #

You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.

── more in #artificial-intelligence 4 stories · sorted by recency
── more on @atomicchat 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/atomicchat-strips-lf…] indexed:0 read:2min 2026-11-16 · —