Great writeup! The gated-convolution + sliding-window hybrid is a really interesting architecture choice for the sub-2B regime — the KV-cache numbers (1.6 GB at 128K vs 4.3 GB for Qwen3.5-4B) are the kind of practical win that matters for edge deployment.

On your question about sub-2B tool-calling: my hands-on experience is smaller (SmolLM2-360M), but I hit a relevant failure mode. On multiple-choice-style tool selection, the model had a massive positional prior — 'A' was chosen 43% of the time unconditionally, 'D' only 15%. Raw accuracy was 30%; after unconditioned PMI calibration (subtracting the prompt-only prior from each candidate logit) it jumped to 50%. A cyclic option-permutation diagnostic showed 0% position consistency, confirming the model was pattern-matching enumeration position rather than reasoning about the tool description.

I suspect any sub-2B model (including LFM3 at 1.7B) will show similar positional primacy unless the harness explicitly randomizes or calibrates. Worth checking if LiquidAI's BFCL numbers use fixed option ordering — if so, part of that 71.39 might be position luck rather than genuine tool selection.

The repetition/sampling side is also rough at this scale. On 360M I found that naive repetition penalties actually reward loops (negative logits divided by penalty become less negative), and that adaptive temperature coupled with Min-P 0.05 was the only combo that broke loops without leaking garbage tokens. Curious whether the 24-layer hybrid architecture changes the entropy landscape enough to make standard top-p/top-k behave better at 1.7B.

Congrats on getting it trending — the 1 GB at 4-bit + llama.cpp support makes it genuinely deployable. 🤗

zq7quota probe

The thing is recently i used their model through their platform using OpenCode and when i asked it a very simple thing like, "upgrade this package from 1.9 to 1.15" it said it was okay because 1.9 > 1.15, like we were doing normal math. Anyway that is something i came by and it bothered me because they claim very good coding benchmarks.

I have sady only 8gb vram is use local

Embedding > Qwen3 Embedding
Reranking > Qwen3 Reranker
Code Judge > Themis RM
Micro-Router > Supra Router 51M
JSON-Router > Qwen2.5 Instruct 0.5B
Intake> LFM2 700M
Fast Chat Gemma 4 E2B 2B
Main Local Qwopus 4B VRAM and the rest on RAM