Results / LFM2.5-24B-A2B Q4_K_M (llama.cpp, expert offload)
LFM2.5-24B-A2B Q4_K_M (llama.cpp, expert offload)
mswea-llamacpp-lfm25llamacppomarchy (RTX 3060 12 GB)steered
The original prompt had no render check, so the first pass of this build was never checked for things like see-through blocks. We added the check (checklist 1.1.0) after the run, re-checked the top builds, and resumed this same session with plain-words requests about what the check found (see Steers below). Wall time covers both parts; the other numbers on this page are for the resume.
The game opens in a new tab. It runs sandboxed, so a game that saves to browser storage may not keep its save.
Metrics
- Checklist score
- 25 / 100
- Design (rubric, 0–5)
- –
- Stop reason
- killed
- Model time
- 1hr 11mins
- Tool time (wall minus model)
- 1hr 19mins
- Wall time
- 2hr 30mins
- Tool calls
- 80
- Tool errors
- 3
- Errors by type
- {"other": 1, "timeout": 2}
- Loop-guard events
- 0
- Tokens in / out
- 2,844,458.0 / 25,589.0
- Thinking tokens
- –
- Decode / prefill tok/s
- 23.2 / 418.7
- Prefix reuse
- 0.986
- Peak VRAM (GPU / engine)
- 10,888 / 10,880 MiB
- Peak engine RAM
- 14,494 MiB
- Engine
- llamacpp
- Model
- LFM2.5-24B-A2B Q4_K_M (llama.cpp, expert offload) (lfm25-24b-a2b)
- Settings
- {"thinking": "always on (LFM2.5 template opens <think>), server --reasoning-budget 3072", "reasoning_budget_tokens": 3072, "max_tokens": 7168, "sampling": "LFM2.5 card: temp 0.1, top_k 50, repeat_penalty 1.05 (no top_p/min_p)", "context": "65536 (q8_0 KV, --fit on expert offload), 32768 fallback", "tool_format": "LFM2 pythonic, parsed by llama.cpp (--jinja)", "research": "research/lfm2.5-24b-a2b.md"}
- Node
- omarchy (RTX 3060 12 GB)
- Prompt
- creative-v3
- Started
- 2026-10-17T04:29:39Z
Feature scores
- O
- 0.75
- C1
- 0.5
- C4
- 0
- C2
- 0
- C3
- 0
- C5
- 0
Screenshots
Landing page at desktop and phone widths, and the game at noon, at night, after breaking a block, and on a phone.
Checklist (3 pass, 3 partial, 12 fail)
| Feature | Test | Score | Evidence |
|---|---|---|---|
| O | files_present | 0.5 | {"required": ["index.html", "game.html"], "notes_md": false} |
| C1 | landing_loads | 0 | {"status": 200, "errors": ["Cannot read properties of undefined (reading 'toDataURL')"], "network_failed": []} |
| C1 | hero_preview | 0.5 | {"std": 8.7, "frame_diff": 0, "size": [1280, 720]} |
| C1 | gallery | 0 | {"images": 3, "generated_nonblank": 0, "distinct": 1, "detail": []} |
| C1 | landing_content | 1 | {"h1": "Crystal Coliseum", "features": 5, "controls": 100, "play": true} |
| C1 | play_now | 0.5 | {"url": "/game.html", "game_ready": false} |
| C1 | responsive_mobile | 1 | {"no_horizontal_scroll": true, "play_now_fits": true, "title_fits": true, "inner": 390, "scroll": 390, "play": [0, 158, 158], "h1": [0, 390, 390]} |
| C4 | game_loads | 0 | {"exception": "Waiting failed: 25000ms exceeded"} |
| C2 | canvas_renders | 0 | {"skipped": "game did not load"} |
| C2 | fps | 0 | {"skipped": "game did not load"} |
| C2 | input_responds | 0 | {"skipped": "game did not load"} |
| C2 | game_states | 0 | {"exception": "Waiting failed: 25000ms exceeded"} |
| C2 | deterministic | 0 | {"exception": "Waiting failed: 25000ms exceeded"} |
| C3 | stays_smooth | 0 | {"exception": "Waiting failed: 25000ms exceeded"} |
| C2 | touch | 0 | {"exception": "Waiting failed: 25000ms exceeded"} |
| C5 | ground_contact | 0 | {"exception": "Waiting failed: 25000ms exceeded"} |
| C5 | render_stable | 0 | {"exception": "Waiting failed: 25000ms exceeded"} |
| O | self_contained | 1 | {"no_external_requests": true, "pinned_three": true, "no_third_party_branding": true, "external": [], "branded_files": []} |
Suite c1.1.2 · Chrome/152.0.7977.82 · WebGL: ANGLE (Intel, Vulkan 1.4.354 (Intel(R) UHD Graphics 630 (CML GT2) (0x00009BC8)), Intel open-source Mesa driver)
Steers (1)
#None · · start · score before
Play now opens the game page, but the game never starts, it just sits there. The front page still shows an error and the three pictures are the same blank image. Please make the game itself start first, even a very simple version where I can move, then fix the front page error and the pictures.


