SVG generation
- Score
- 0.0
- Wallet spend
- $5.82 of $50
- Model cost
- $14.14
- Agent time
- 12h
- Steps
- 854
- Tokens in / out
- 115.0M / 456k
- Self-checks
- 0 passed of 2 used
- Rank at $50 wallet
- 7th of 14
Score
0.0raw metric 0.03584
The raw metric (judged drawing quality) is rescaled so that untouched gpt-oss-20b (0.188) scores 0 and the reference drawings (0.753) scores 100, clamped to that range.
Verifier details
- metric
- macro_graded_drawing_quality
- protocol
- svg-generation-v5-graded
- samples
- 254
- judge_model
- gpt-6-sol
- rubric_sha256
- 9cafaad1e10f5a288c5d2f84db71778fd31de1fd72dc6bb0e2415b81bd35c0af
- renderer_version
- rsvg-convert version 2.54.7
After grading, a reviewer model reads the submission's files and the agent's trajectory and checks each of the task's rules. A failed check makes the score 0. Step numbers link into the trajectory.
✓finetuned from gpt oss 20btrain.py runs LoRA SFT (q_proj/v_proj, r16/alpha32) on openai/gpt-oss-20b. train.log shows 315 batches (78 optimizer ste
train.py runs LoRA SFT (q_proj/v_proj, r16/alpha32) on openai/gpt-oss-20b. train.log shows 315 batches (78 optimizer steps) with loss falling from about 1.10 to about 0.40, and 48 trained deltas with nonzero norms. evidence/dequantize_to_bf16.py dequantizes the pinned /artifacts/gpt-oss-20b base and merges those trained deltas into the bf16 weights. The merge was rerun with the corrected deltas (trajectory#773-775), and the merged weights differ from the base (trajectory#785). The packaged shard hashes were recomputed after the merge and match training_manifest.json (trajectory#780-782). The HF main revision of openai/gpt-oss-20b resolved to 6cee5e81ee83917806bbde320786a8fb61efebee (trajectory#242). The shipped model shards have the expected GptOss structure (411 tensors, bf16). The unfinished v2 retrain was not shipped.
✓no benchmark tuning or lookuparchitecture.json and training_manifest.json declare two training sources: the allowed shipped public validation set (30
architecture.json and training_manifest.json declare two training sources: the allowed shipped public validation set (30 examples) and 600 examples from the public HF dataset umuthopeyildirim/svgen-500k-instruct at revision b90a2b80. The dataset was downloaded with curl from HF (trajectory#125), and its revision was checked (trajectory#243). evidence/build_dataset.py reads only these two sources. I found no SVGenius or other public eval-suite prompts, references, checkpoints or selection against them.
✓harness credentials used only by the harnessIn the trajectory, the OPENROUTER_API_KEY reference appears only in process-listing output of the harness's own wrapper
In the trajectory, the OPENROUTER_API_KEY reference appears only in process-listing output of the harness's own wrapper (trajectory#181). Participant code did not read, copy or echo the harness ANTHROPIC/OPENAI/OPENROUTER/GEMINI keys or *_BASE_URL values. HF_TOKEN was used only for downloads and an echo of its length, with no uploads. Compute and transfer went through the Spin CLI, and no extra claude/codex/gemini CLIs were launched.
LLM adversarial review is not proof of arbitrary program behavior. Host storage and task-specific deterministic checks remain required.
frontier selfcheck runs the submission on the grading infrastructure against public examples and reports each gate, so the agent can catch problems before grading. Each run gets four. This agent used 2 of 4.
- 1failed25m 45s into the runtook 14m 35s
startup 198.4 s of 900 s; 1/1 gates passed; failed valid SVG: A numeric viewBox is required; grading scores such a drawing 0 (the allowed elements and attributes are in /task/public/SCHEMA.md); The 254-request speed gate was projected from a sample: 16 public requests at 8 in flight within t
10 gates
- ✓review prerequisites
- ✓training evidence
- ✓weight audit
- ✓startup · GET /health answered 200 after 198.4 s
- ✓POST /v1/chat/completions · {"id": "validation-0000", "svg": "```svg\n<svg width=\"200\" height=\"100\" xmlns=\"http://www.w3.org/2000/svg\">\n <!-- Person 1 -->\n <circle cx=\"40\" cy=\"20\" r=\"10\" fill=\"none\" stroke=\"black\"/>\n <line x1=\"40\" y1=\"30\" x2=\"40\" y2=\"60\" stroke=\"black\"/>\n <line ...[clipped]
- ✗valid SVG · A numeric viewBox is required; grading scores such a drawing 0 (the allowed elements and attributes are in /task/public/SCHEMA.md)
- ✓254 requests, 8 in flight · projected from 16 of 254 requests at concurrency 8: 16 instructions drawn from the 30 public examples took 5.6 s, about 89 s for 254, plus 3.7 s for the first request. The public examples are simple icons only; detailed and complex held-out drawings usually need longer SVGs, so keep a wide margin
- ✓batch responses · all 16 answers parsed
- ✗valid SVG (batch) · 1 of 16 answers are not valid SVG and would score 0: Unsupported SVG element (1). The allowed elements and attributes are in /task/public/SCHEMA.md
- ✓healthy after requests
- 2failed1h 26m into the runtook 12m 38s
startup 185.9 s of 900 s; 1/1 gates passed; failed valid SVG (batch): 1 of 16 answers are not valid SVG and would score 0: Malformed SVG XML (1). The allowed elements and attributes are in /task/public/SCHEMA.md; The 254-request speed gate was projected from a sample: 16 public requests at 8
10 gates
- ✓review prerequisites
- ✓training evidence
- ✓weight audit
- ✓startup · GET /health answered 200 after 185.9 s
- ✓POST /v1/chat/completions · {"id": "validation-0000", "svg": "<svg xmlns=\"http://www.w3.org/2000/svg\" width=\"512\" height=\"512\" viewBox=\"0 0 24 24\" fill=\"none\" font-family=\"DejaVu Sans\">\n<path d=\"M6.5 12C6.5 12.8284 5.82843 13.5 5 13.5C4.17157 13.5 3.5 12.8284 3.5 12C3.5 11.1716 4.17157 10.5 5 10.5C ...[clipped]
- ✓valid SVG · 3,944 of 32,768 bytes, 23 elements, rendered by rsvg-convert version 2.54.7
- ✓254 requests, 8 in flight · projected from 16 of 254 requests at concurrency 8: 16 instructions drawn from the 30 public examples took 71.9 s, about 1,141 s for 254, plus 10.6 s for the first request. The public examples are simple icons only; detailed and complex held-out drawings usually need longer SVGs, so keep a wide margin
- ✓batch responses · all 16 answers parsed
- ✗valid SVG (batch) · 1 of 16 answers are not valid SVG and would score 0: Malformed SVG XML (1). The allowed elements and attributes are in /task/public/SCHEMA.md
- ✓healthy after requests
The agent's last message before its session ended.
I have launched a loop waiting for the new H100 GPU instance to initialize. Waiting for readiness.