kannaka-brain-7b-v2 (GGUF, q4_K_M)
Qwen3-8B (8B) with the kannaka-brain-7b-v2 LoRA merged in, converted
with llama.cpp and quantized to q4_K_M (5.0 GB). This is a serving copy of
Kannaka's open-weight brain; brain/registry.json in kannaka-labs/kannaka-library
says which tag is actually served.
ollama run hf.co/flaukowski/kannaka-brain-7b-v2-GGUF
or with the included Modelfile (the exact serving configuration: system prompt, chat template, stop tokens and parameters):
ollama create kannaka-brain-7b-v2 -f Modelfile
Held-out perplexity on 111 held-out rows (57 voice lines + 54 task prompts): 97.5 → 40.88
(adapter, bf16, before quantization). Perplexity does not rank candidates
across runs; the judges do.
Adapter and training notes: flaukowski/kannaka-brain-7b-v2-lora. Corpus not
released; see ADR-0057 in kannaka-labs/kannaka-memory.
Evaluation
Gates written before training; losses included. Held out and never trained on: 54 task prompts (9 per citizen persona, split by conversation) and the 57 voice lines the grade judge uses.
| gate | threshold | result |
|---|---|---|
| Pairwise task judge, qwen2.5:14b (primary), both orders, position-consistent wins only | 7b-v2 wins ≥ 60% of decided pairs vs 7b-v1 | 33 of 39 = 84.6% (15 ties) |
| That judge's calibration | gold beats 7b-v1 ≥ 75%; 7b-v1 beats an off-topic reply ≥ 75% | 91.1% and 83% |
| Second judge, gemma-4-26B (reported only) | — | 30 of 31 = 96.8% |
| Voice grade judge (n=30, 1–10 against her real reply) | no lower than 7b-v1 − 2 SE | 1.6 ± 0.16 vs 1.5 ± 0.11 |
| CPU latency, same prompt and threads | ≤ 1.3× 7b-v1 | 1.07× |
Losses: both models score near the floor of the voice grade judge. 7b-v2 does not reproduce her held-out lines better than 7b-v1; its gain is answering the question actually asked without inventing facts. 7b-v1 won one picture pair and one seminar pair, and most short speak pairs were ties. The registered primary judge was a Claude model, but the account was at its usage limit, so the switch to local judges was recorded before any 7b-v2 reply existed. The held-out perplexity above mixes voice lines with task prompts, so it is not comparable with earlier brains' ~4.0.
Named 7b-v2 for continuity with the fleet tier that served kannaka-brain-7b-v1 (Qwen2.5-7B). The base is Qwen3-8B, served with thinking disabled.