ModelMOD-2026-0006v0.1

Qwythos-9B-v2 (Q4_K_M, local, Ollama)

Open-weight 9.0B model of the qwen35 family, GGUF Q4_K_M, served locally by Ollama 0.33.3 on an RTX 3070. Used as generator, selector and judge in the real local run of the PACC-Hybrid v0.2 protocol, with thinking disabled in runs 1–2 and enabled (effort medium, +3,500 output tokens per call) in run 3.

Research status: STABLE the current conclusions are relatively stable
Evidence level: E2 Controlled experiment
Data basis: REAL MODEL A real language model was executed; the model, its version and its configuration are recorded on the page.
Version: 0.1
Updated: 2026-09-11
Created: 2026-09-11
Domain: Evaluation
Program: PRG-2026-0001 Adaptive Epistemic Systems
Authors: Neo.K (EveMissLab)
AI collaborators: Sol (GPT-5.6, OpenAI ChatGPT)

Record

provider: empero-ai (Hugging Face GGUF) via Ollama
model_version: 0881a8d857f619cffa33395e9a1c3d1e49eb512598bac4674ed3499bdc308114
access_type: local, loopback only; open weights
context_window: 1048576
configuration_notes: run tag qwythos-9b-v2-q4km-ctx8k:latest = base hf.co/empero-ai/Qwythos-9B-v2-GGUF:Q4_K_M (digest 5008e78bba127262f3f7ad86425bb49a5e0f47bb1959a4d30bfe17832ec45856) + PARAMETER num_ctx 8192 via scripts/Modelfile.qwythos-ctx8k; Ollama's default 4096 context aborted the first primary attempt after 55 calls
reasoning.effort = none (thinking off) for every call
Ollama /v1/responses, OpenAI-compatible; package provider unchanged
run 3 tag qwythos-9b-v2-q4km-ctx12k:latest = same base + PARAMETER num_ctx 12288 (digest 7ffdc602f28799ffa312ea8dc85c364b047e1386380c4616c21b02c96c3f85d5); served with OLLAMA_FLASH_ATTENTION=1 and OLLAMA_KV_CACHE_TYPE=q8_0; reasoning.effort = medium with +3500 output tokens on every call
known_behavior_notes: With thinking enabled, hidden reasoning consumes the protocol's output-token budgets and output_text comes back empty.
As judge it sometimes returns a rubric sentence as pattern_label, which makes pattern-entropy numbers fragile.
With thinking on and +3,500 tokens per call the protocol runs (380 calls, 3.7 h at 25–33 tok/s); reasoning still exhausted the whole budget once in 380 calls (empty output, regenerated). All deterministic literal checks pass with thinking on.
canonical_external_reference: https://huggingface.co/empero-ai/Qwythos-9B-v2-GGUF

Relations

Source	Relation	Target	Status	ID
`SYS-2026-0003` PACC-LLM Hybrid Lab	`uses_model`	`MOD-2026-0006` Qwythos-9B-v2 (Q4_K_M, local, Ollama)	`ACTIVE`	`REL-2026-0325`
`EXP-2026-0023` PACC-Hybrid v0.2 — first real-model run, on a local 9B open-weight model	`uses_model`	`MOD-2026-0006` Qwythos-9B-v2 (Q4_K_M, local, Ollama)	`ACTIVE`	`REL-2026-0328`
`EXP-2026-0024` PACC-Hybrid v0.2 — real-model run 2: four repetitions and label-free creative breadth	`uses_model`	`MOD-2026-0006` Qwythos-9B-v2 (Q4_K_M, local, Ollama)	`ACTIVE`	`REL-2026-0336`
`EXP-2026-0025` PACC-Hybrid v0.2 — real-model run 3: thinking enabled with enlarged output budgets	`uses_model`	`MOD-2026-0006` Qwythos-9B-v2 (Q4_K_M, local, Ollama)	`ACTIVE`	`REL-2026-0345`
`EXP-2026-0103` XA-06 first real-model pilot — Qwythos-9B-v2 on the A0→A5 ladder (36 trials, 2026-09-03)	`uses_model`	`MOD-2026-0006` Qwythos-9B-v2 (Q4_K_M, local, Ollama)	`ACTIVE`	`REL-2026-0474`

History and provenance

Canonical URL

https://evemisslab.com/ai/models/MOD-2026-0006/

Machine-readable

/ai/models/MOD-2026-0006/index.json

Snapshot

AI-SNAPSHOT-v0.1-fe85b9694a45

Provenance

source: EveMissLab research collection: Adaptive Epistemic Systems (真本體論13)
extracted_by: Splice (Claude Code), reading the canonical UTF-8 sources and each lab's own result reports
extracted_at: 2026-09-11
generator: tools/extract_aes/extract.py
claim_boundary: status, evidence level and result type follow the source artifact's own stated claim boundary; nothing is upgraded beyond what the report supports