dpsk-v4-flash.h008.keep_sft1.step_700

AgentPTB sweep checkpoint. Cell dpsk-v4-flash — pi / DeepSeek v4-flash @ effort thinking.

field	value
plot cell	dpsk-v4-flash
driver	pi / DeepSeek v4-flash
reasoning effort	thinking
run boot (UTC)	2026-08-11T08:38:35Z
role	intermediate
hours into run	h8.26 of 100
checkpoint path in run	ckpts/keep_sft1/step_700
shards	1
size	18.8 GB
base model	Qwen/Qwen3.5-9B-Base
eos_token_id	[248044, 248046] ✅ correct (repaired 2026-08-28)

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.

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.

Packaging repair (2026-08-28)

generation_config.json now sets eos_token_id to [248044, 248046], so generation stops on both <|endoftext|> and the chat-format <|im_end|> token. Model weights are unchanged.