dsh-plain-talk 0.1.9 → 0.1.10
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +4 -0
- package/README.en.md +7 -5
- package/README.md +7 -5
- package/package.json +1 -1
package/CHANGELOG.md
CHANGED
package/README.en.md
CHANGED
|
@@ -43,12 +43,14 @@ Environment: DeepSeek Harness `desktop` profile, headless run, model `command co
|
|
|
43
43
|
|
|
44
44
|
|Case|Characters (off / on)|Output tokens (off / on)|
|
|
45
45
|
|-|-|-|
|
|
46
|
-
|A1 Create and verify files|
|
|
47
|
-
|A2 Write and read back|
|
|
48
|
-
|B1 Why models forget|
|
|
46
|
+
|A1 Create and verify files|139 → 184, 32.37%↑|2,333 → 1,344, 42.39%↓|
|
|
47
|
+
|A2 Write and read back|80 → 68, 15.00%↓|220 → 398, 80.91%↑|
|
|
48
|
+
|B1 Why models forget|408 → 348, 14.71%↓|732 → 326, 55.46%↓|
|
|
49
49
|
|B2 merge vs rebase|184 → 194, 5.43%↑|150 → 366, 144.00%↑|
|
|
50
|
-
|C1 Experiment draft|
|
|
51
|
-
|C2 Release-check draft|
|
|
50
|
+
|C1 Experiment draft|518 → 453, 12.55%↓|2,781 → 3,066, 10.25%↑|
|
|
51
|
+
|C2 Release-check draft|645 → 324, 49.77%↓|1,636 → 2,092, 27.87%↑|
|
|
52
|
+
|
|
53
|
+
Token counts cover every model step in the turn (including reasoning and tool calls), so they do not always move with reply length; C2, for example, is 321 characters shorter yet spends more tokens on analysis and self-checking.
|
|
52
54
|
|
|
53
55
|
About +589 input token per step.
|
|
54
56
|
|
package/README.md
CHANGED
|
@@ -43,12 +43,14 @@ DeepSeek Harness 的回复优化插件：输入框里一个「说人话」开关
|
|
|
43
43
|
|
|
44
44
|
|场景|字数（关闭 / 开启）|输出 token（关闭 / 开启）|
|
|
45
45
|
|-|-|-|
|
|
46
|
-
|A1 建文件并验证|
|
|
47
|
-
|A2 写入并读回|
|
|
48
|
-
|B1 为什么模型会忘事|
|
|
46
|
+
|A1 建文件并验证|139 → 184, 32.37%↑|2,333 → 1,344, 42.39%↓|
|
|
47
|
+
|A2 写入并读回|80 → 68, 15.00%↓|220 → 398, 80.91%↑|
|
|
48
|
+
|B1 为什么模型会忘事|408 → 348, 14.71%↓|732 → 326, 55.46%↓|
|
|
49
49
|
|B2 merge 与 rebase|184 → 194, 5.43%↑|150 → 366, 144.00%↑|
|
|
50
|
-
|C1 实验草稿|
|
|
51
|
-
|C2 上线检查草稿|
|
|
50
|
+
|C1 实验草稿|518 → 453, 12.55%↓|2,781 → 3,066, 10.25%↑|
|
|
51
|
+
|C2 上线检查草稿|645 → 324, 49.77%↓|1,636 → 2,092, 27.87%↑|
|
|
52
|
+
|
|
53
|
+
token 统计覆盖整轮所有模型步（含思考与工具调用），因此与最终回复字数不必然同向；例如 C2 最终回复少了 321 字，但开启后多花了分析与自检的 token。
|
|
52
54
|
|
|
53
55
|
每步约 +589 input token。
|
|
54
56
|
|