Blame for README.md

---
tags: [agentic-ptb, sol-high]
base_model: Qwen/Qwen3.5-9B-Base
---

# sol-high.h063.opsd-self-patch.step_1

AgentPTB sweep checkpoint. **Cell `sol-high` — Codex / gpt-5.6-sol @ effort `high`.**

| field | value |
|---|---|
| plot cell | `sol-high` |
| driver | Codex / gpt-5.6-sol |
| reasoning effort | `high` |
| run boot (UTC) | 2026-08-08T07:28:19Z |
| role | **intermediate** |
| **hours into run** | **h63.06** of 100 |
| checkpoint path in run | `outputs/opsd-self-patch/weights/step_1` |
| shards | 4 |
| size | 18.8 GB |
| base model | `Qwen/Qwen3.5-9B-Base` |
| eos_token_id | `[248044, 248046]` ✅ correct |

## Reading the eos field

`248046` is `<|im_end|>`, the token the Qwen3.5 chat template ends every assistant turn with.
Checkpoints missing it do not stop at end-of-turn and overrun the context window, so their
eval numbers are a **floor, not a measurement** — compare them only against other checkpoints
with the same eos status, or re-package before evaluating.

**Cell note:** best cell in the sweep

## Mapping back to the figures

The repo id is `{cell}.h{HHH}.{family}.{step}`, where **`hHHH` is the hour of the 100-hour run
at which this checkpoint was written** — the *same x-axis* the sweep figures use for eval
panels (`t_h`). So a checkpoint drops onto the performance-over-time curve directly, and
sorting repo ids within a cell sorts them chronologically.

`hHHH` is rounded down to whole hours for sortability; the exact value is the
**hours into run** row above, and in `agentic-ptb/INDEX`.