docs(solutions): CJK token estimation undercount fix
Test / backend-test (pull_request) Has been cancelled Details
Test / frontend-unit (pull_request) Has been cancelled Details
Test / api-e2e (pull_request) Has been cancelled Details
Test / frontend-e2e (pull_request) Has been cancelled Details

Document the ContextCompressor CJK 4x underestimation bug and fix:
- estimate_text_tokens() CJK 1:1 / ASCII 4:1 heuristic
- _summarize() max_chars budget fix (P1: was * 4, allowed 4x CJK budget)
- Linear compress flow + structured logging
- Prevention: charset-aware heuristics, audit dependent truncation points
This commit is contained in:
chiguyong 2026-07-03 09:40:09 +08:00
parent 3a05c4d1e6
commit 027f7909aa
1 changed files with 152 additions and 0 deletions

View File

@ -0,0 +1,152 @@
---
title: "ContextCompressor CJK token estimation undercounted by 4x"
date: 2026-07-03
module: core/compressor
component: assistant
tags:
- cjk
- token-estimation
- context-compression
- react-engine
- heuristic
problem_type: logic_error
severity: high
symptoms:
- "estimate_tokens() uses len(content) // 4 (ASCII heuristic), undercounting CJK tokens by ~4x since CJK chars are ~1 char per token"
- "Context compression triggers too late for CJK-heavy conversations, risking context window overflow"
- "_summarize() pre-truncation uses max_chars = max_input_tokens * 4, allowing CJK text to send 4x the token budget to the LLM"
- "ReActEngine._should_compress() fallback inherits the same flawed len // 4 estimation for compressors without should_compress()"
root_cause: logic_error
resolution_type: code_fix
---
## Problem
`ContextCompressor.estimate_tokens()` 使用 `len(content) // 4`ASCII 启发式4 字符 ≈ 1 token估算 token 数。对 CJK中文/日文/韩文文本而言1 个字符 ≈ 1 token因此该估算器将 CJK token 数低估约 4 倍。这导致 CJK 为主的会话压缩触发过晚,存在上下文窗口溢出风险。`_summarize()` 的预截断 `max_chars = max_input_tokens * 4` 进一步放大了问题——允许 CJK 文本向 LLM 发送 4 倍 token 预算的输入。
## Symptoms
- CJK 为主的会话在远超预期阈值(约 4 倍)后才触发压缩,`should_compress()` 返回 `False` 而实际 token 已超出 `model_context_limit`
- `_summarize()` 可能向 LLM 发送 4 倍 token 预算的 CJK 文本P1 缺陷——可能触发上下文上限 / 400 错误)
- 中文长会话面临上下文窗口溢出 / 请求失败的高风险
- 压缩流程依赖递归深度 `_compression_depth`,难以观测与调试,缺乏结构化日志
## What Didn't Work
旧的 `len(content) // 4` 估算器基于 ASCII/拉丁语平均比例(约 4 字符/token。对纯 CJK 文本1 字符 ≈ 1 token该估算产生约 4 倍偏低的估计。例如4000 个中文字符会被估算为 1000 token但实际约 4000 token。这导致
- `should_compress()` 的 headroom 阈值(`model_context_limit * 0.8`)直到实际 token 已 4 倍超出预期阈值才触发
- `_summarize()``max_chars = max_input_tokens * 4` 在预算为 3200 token 时放行 12800 字符≈12800 CJK token— 此 P1 缺陷由 ce-code-review 捕获 (session history)
- 旧的递归式 `compress()` 依赖 `_compression_depth` 计数器,流程难以阅读,且 `_compress_aggressive()` 接收已压缩的 `compressed` 列表,存在 summary-of-summaryF-010风险
## Solution
**1. 新增 CJK 感知的 token 估算器**`src/agentkit/core/compressor.py`
```python
def _is_cjk(char: str) -> bool:
"""Check if a character is CJK (1 token ≈ 1 char)."""
cp = ord(char)
return (
0x4E00 <= cp <= 0x9FFF # CJK Unified Ideographs
or 0x3040 <= cp <= 0x30FF # Hiragana + Katakana
or 0xAC00 <= cp <= 0xD7AF # Hangul Syllables
)
def estimate_text_tokens(text: str) -> int:
"""Estimate token count: CJK 1:1, other characters 4:1."""
cjk_count = 0
non_cjk_count = 0
for char in text:
if _is_cjk(char):
cjk_count += 1
else:
non_cjk_count += 1
return cjk_count + non_cjk_count // 4
```
`estimate_tokens()` 改用 `estimate_text_tokens()`
```python
def estimate_tokens(self, messages: list[dict]) -> int:
"""Estimate total tokens in message list (CJK 1:1, ASCII 4:1)"""
return sum(estimate_text_tokens(str(m.get("content", ""))) for m in messages)
```
**2. 修复 `_summarize()` 预截断**P1 缺陷,由 ce-code-review 捕获):
```python
# 修改前CJK 可 4 倍超预算):
max_chars = max_input_tokens * 4
# 修改后CJK 1:1 精确ASCII 4:1 保守):
max_chars = max_input_tokens
conversation_text = conversation_text[:max_chars] + "\n...[truncated]"
```
**3. `ReActEngine._should_compress()` 回退路径同步使用 `estimate_text_tokens`**`src/agentkit/core/react.py` 约 1750 行):
```python
# Fallback: fixed threshold for compressors without headroom support
estimated_tokens = sum(
estimate_text_tokens(str(m.get("content", ""))) for m in conversation
)
return estimated_tokens > self._DEFAULT_COMPRESS_THRESHOLD
```
**4. 重写 `compress()` 为线性流程**,移除递归式 `_compression_depth`
```python
async def compress(self, messages: list[dict]) -> list[dict]:
"""Linear flow: summarize -> aggressive -> truncate."""
tokens_before = self.estimate_tokens(messages)
if tokens_before <= self._max_tokens:
return messages
# ... 分离 system/old/recent ...
# Step 1: summarize
# Step 2: aggressive (F-010: 传入原始 messages 而非 compressed避免 summary-of-summary)
if self.estimate_tokens(compressed) > self._max_tokens:
compressed = await self._compress_aggressive(messages)
# Step 3: truncate as last resort
if self.estimate_tokens(compressed) > self._max_tokens:
compressed = self._truncate(compressed)
# Step 4: 结构化日志
self._log_compression(tokens_before, tokens_after, len(messages), len(compressed), strategy)
return compressed
```
**5. 新增 `_log_compression()` 结构化日志**
```python
logger.info(
"context compressed: %d -> %d tokens (%.1f%%), messages: %d -> %d, strategy: %s",
tokens_before, tokens_after, ratio * 100,
msg_count_before, msg_count_after, strategy,
)
```
## Why This Works
**根本原因**CJK 字符在主流 tokenizerBPE/WordPiece/SentencePiece中近似 1:1 映射为 token而 ASCII/拉丁文约 4 字符/token。`len(content) // 4` 把这 4 倍差异抹平了,导致 CJK 估算系统性偏低。
修复后的 `estimate_text_tokens()` 对 CJK 字符按 1:1 计数、对非 CJK 保留 4:1既纠正了 CJK 偏差又维持了 ASCII 行为。`_summarize()` 的 `max_chars = max_input_tokens` 对 CJK 精确1:1、对 ASCII 保守(截断到 1/4 预算但安全),彻底消除了"4 倍超预算"路径。
`headroom_threshold=0.8` 吸收了纯 CJK 仍可能存在的 10-20% 估算偏差(`ponytail:` 注释已标注上限与升级路径——`litellm.token_counter` 或 provider 专用 tokenizer
**可维护性改进**:线性 `compress()` 流程summarize → aggressive → truncate移除了递归深度计数器单次读取即可理解全部降级路径`_compress_aggressive()` 接收原始 `messages` 而非已压缩的 `compressed`,规避了 F-010 的 summary-of-summary`_log_compression()` 提供结构化观测字段before/after/ratio/msg_count/strategy使压缩行为可调试、可告警。
## Prevention
- **字符集感知的估算启发式**:任何 token 估算逻辑必须考虑字符集差异。`len // 4` 仅对 ASCII 成立CJK/emoji/其他多字节脚本需分别处理。涉及多语言输入时,优先使用 `litellm.token_counter` 或 provider 专用 tokenizer当前 `estimate_text_tokens``ponytail` 注释已标注此升级路径)
- **修改估算逻辑时审计所有依赖的截断点**:本次修复同步审计了 `estimate_tokens()`、`_summarize()` 预截断、`ReActEngine._should_compress()` 回退路径。`_truncate()` 仍用 `len(content) > self._max_tokens * 4` 判断OQ21 延期项),后续需同步迁移至字符集感知逻辑
- **测试覆盖字符集矩阵**:新增测试覆盖纯 CJK、纯 ASCII、CJK+ASCII 混合、平假名/片假名、谚文等场景(`tests/unit/test_context_compressor.py`),验证估算器在各字符集下的正确性
- **避免 summary-of-summary**多级压缩时后续阶段应接收原始输入而非前级压缩产物防止信息逐级失真F-010 教训)
- **结构化日志先行**:压缩是黑盒操作,必须输出 before/after token、压缩比、消息数、策略等结构化字段便于线上问题定位
## References
- Plan: `docs/plans/2026-07-02-003-feat-context-compressor-cjk-prefix-enhancement-plan.md`3 轮 ce-doc-review收敛到 U1 + U3
- PR #21: `feat/context-compressor-cjk`commits `be45fe4` + `3a05c4d`
- Upstream context: `docs/plans/2026-06-24-004-feat-long-horizon-reliability-optimization-plan.md`headroom 压缩引入点)
- Residual: OQ21`_truncate()` `* 4` 一致性P2 manual 跟进)