fix(stats): TPM now only counts output tokens

[Soulter]

改动内容

问题

TPM（平均每分钟词元数）之前统计的是总 token（input + output），但作为生成性能指标，应该只统计 output tokens。

改动

• 后端 (stat.py): 新增 range_total_output_tokens，仅累加 record.token_output，TPM 公式改为使用 output tokens
• 前端国际化: 标签明确为输出词元数
  • zh-CN: 平均每分钟输出（TPM）
  • en-US: Average Output TPM
  • ru-RU: Средний Output TPM

影响

• ✅ 仅 TPM 指标改为 output-only
• ✅ 其他统计（趋势图、Provider/UMO排名、今日统计等）保持原样不变

Summary by Sourcery

Update token statistics so that TPM is calculated using only output tokens and align UI labels accordingly.

Bug Fixes:

• Correct TPM calculation to use only output tokens instead of total input+output tokens.

Enhancements:

• Add a dedicated accumulator for output tokens in provider token statistics to support accurate TPM computation.

Documentation:

• Clarify stats UI labels in all supported locales to indicate that TPM reflects average output tokens per minute.

[sourcery-ai]

Hey - I've left some high level feedback:

• Consider guarding range_total_output_tokens += record.token_output against record.token_output being None or missing to avoid potential runtime errors when older or malformed records are processed.
• Now that range_avg_tpm is output-only, it might be helpful to rename range_total_tokens to something like range_total_io_tokens (or similar) where it is still used, to avoid confusion between total tokens and output-only tokens in future maintenance.

Prompt for AI Agents

Please address the comments from this code review:

## Overall Comments
- Consider guarding `range_total_output_tokens += record.token_output` against `record.token_output` being `None` or missing to avoid potential runtime errors when older or malformed records are processed.
- Now that `range_avg_tpm` is output-only, it might be helpful to rename `range_total_tokens` to something like `range_total_io_tokens` (or similar) where it is still used, to avoid confusion between total tokens and output-only tokens in future maintenance.

---

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

• X
• Mastodon
• LinkedIn
• Facebook

_{Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.}

[gemini-code-assist]

Code Review

This pull request updates the statistics dashboard to calculate the Average TPM (Tokens Per Minute) based on output tokens instead of total tokens, with corresponding updates to the English, Russian, and Chinese localization files. A review comment points out a potential logic error in the TPM calculation: the total output tokens are accumulated for all records, whereas the total duration used as the denominator only accounts for records with valid start and end times, which could lead to inaccurate averages.

[dosubot]

Related Documentation

1 document(s) may need updating based on files changed in this PR:

AstrBotTeam's Space

pr4697的改动

View Suggested Changes

@@ -904,7 +904,7 @@
   - `range_total_calls`：选定时间范围内的总 API 调用次数
   - `range_avg_ttft_ms`：首个令牌平均响应时间（毫秒）
   - `range_avg_duration_ms`：请求平均持续时间（毫秒）
-  - `range_avg_tpm`：平均每分钟令牌数（TPM）
+  - `range_avg_tpm`：平均每分钟输出令牌数（Output TPM），仅统计输出令牌，不包含输入令牌
   - `range_success_rate`：API 调用成功率（0-1 之间）
   - `range_by_provider`：按提供商分类的令牌消耗排名
   - `range_by_umo`：按统一消息来源（UMO）分类的令牌消耗排名

[Accept] [Decline]

^{How did I do? Any feedback?}