Skip to content

v1.0.1

Choose a tag to compare

View all tags
@zerro-223 zerro-223 released this 15 Aug 01:37
· 10 commits to main since this release
v1.0.1
122c785

Token usage statistics for the DeepSeek Harness Web UI — v1.0.1.

Added

  • Estimated cost card (USD): per-model prices (input / output / cache read / cache write, USD per 1M tokens) configured in the in-panel pricing dialog, persisted to ~/.dsh/storages/token-stats/prices.json.
  • Automatic price sync from modelradar.cn: fetched on boot with backoff retries, refreshed daily, cached to prices-auto.json; manual prices always override.

Changed

  • USD-correct auto prices — modelradar prices 183 of 294 models natively in CNY; auto prices now always use the dataset's per-field *PriceUsdPer1M conversions (native CNY used to be treated as USD, overstating DeepSeek costs ~7x). CNY-sourced rows are tagged auto·CNY in the pricing dialog.
  • Async batched persistence (was one blocking appendFileSync per model call); history loading reads only the tail of usage.jsonl; cache-only rows survive a restart.
  • Cache read/write stat cards removed (DeepSeek and most providers report no cache-write metric); the stat grid spans the 6 remaining cards.
  • Pricing dialog: cache-write price column, model search box, inline errors when the prices API is unreachable or a refresh fails, and basename matching for provider-prefixed model ids.
  • Total-tokens card shows a Chinese-unit approximation (≈万 / ≈亿).

Full changelog: https://github.com/zerro-223/dsh-token-usage/blob/main/CHANGELOG.md

Assets 2
Loading