dahara1 commited on
Commit
dcde4f2
ยท
verified ยท
1 Parent(s): 5e1dc30

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +2 -2
README.md CHANGED
@@ -31,7 +31,7 @@ Although llama.cpp can be used to reduce the size of the file with various quant
31
  - ใƒˆใƒƒใƒ—P๏ผˆ--top_p๏ผ‰: ใ“ใฎๅ€คใ‚’ใ•ใ‚‰ใซไฝŽใ่จญๅฎšใ™ใ‚‹ใ“ใจใงใ€ใƒขใƒ‡ใƒซใŒ่€ƒๆ…ฎใ™ใ‚‹ๅ˜่ชžใฎ็ฏ„ๅ›ฒใ‚’็‹ญใ‚ใ€ใ‚ˆใ‚Šไธ€่ฒซๆ€งใฎใ‚ใ‚‹ใƒ†ใ‚ญใ‚นใƒˆใ‚’็”Ÿๆˆใ™ใ‚‹ใ‚ˆใ†ใซใชใ‚Šใพใ™ใ€‚
32
  - ็”Ÿๆˆใ™ใ‚‹ๅ˜่ชžๆ•ฐ๏ผˆ-n๏ผ‰: ใ“ใฎๅ€คใ‚’ๆธ›ใ‚‰ใ™ใ“ใจใงใ€ใƒขใƒ‡ใƒซใŒ็”Ÿๆˆใ™ใ‚‹ใƒ†ใ‚ญใ‚นใƒˆใฎ้•ทใ•ใ‚’็Ÿญใใ—ใ€ไธ่ฆใช่ฟฝๅŠ ใƒ†ใ‚ญใ‚นใƒˆใฎ็”Ÿๆˆใ‚’้˜ฒใใ“ใจใŒใงใใพใ™ใ€‚-1 = ็„ก้™ๅคงใ€-2 = ๆ–‡่„ˆใŒๆบ€ใŸใ•ใ‚Œใ‚‹ใพใงใ€‚
33
 
34
- ไปฅไธ‹ใฏllama.cppใฎไฝœ่€…(ggerganov)ใซใ‚ˆใ‚‹ๆŽจๅฅจใƒ‘ใƒฉใƒกใƒผใ‚ฟใƒผใงใ™
35
  - -e (ๆ”น่กŒ\nใ‚’ใ‚จใ‚นใ‚ฑใƒผใƒ—)
36
  - --temp 0 (ๆœ€ใ‚‚็ขบ็Ž‡ใฎ้ซ˜ใ„ใƒˆใƒผใ‚ฏใƒณใฎใฟใ‚’้ธๆŠž)
37
  - --repeat-penalty 1.0 (็นฐใ‚Š่ฟ”ใ—ใƒšใƒŠใƒซใƒ†ใ‚ฃใ‚’ใ‚ชใƒ•ใ€‚ๆŒ‡็คบ่ชฟๆ•ดๆธˆใƒขใƒ‡ใƒซใงใ“ใ‚Œใ‚’ใ™ใ‚‹ใฎใฏใ€ๆฑบใ—ใฆ่‰ฏใ„่€ƒใˆใจใฏ่จ€ใˆใชใ„ใ€‚)
@@ -42,7 +42,7 @@ Adjust the following parameters as needed
42
  - Top P (--top_p): Setting this value even lower will narrow the range of words considered by the model and produce more consistent text.
43
  - Number of words to generate (-n): Reducing this value will shorten the length of text generated by the model and prevent the generation of unnecessary additional text. -1 = infinity(default), -2 = until context filled.
44
 
45
- The following are the recommended parameters by the author of llama.cpp(ggerganov)
46
  - -e (escape newlines (\n))
47
  - --temp 0(pick most probable tokens)
48
  - --repeat-penalty 1.0(disable repetition penalty (it's never a good idea to have this with instruction tuned models)
 
31
  - ใƒˆใƒƒใƒ—P๏ผˆ--top_p๏ผ‰: ใ“ใฎๅ€คใ‚’ใ•ใ‚‰ใซไฝŽใ่จญๅฎšใ™ใ‚‹ใ“ใจใงใ€ใƒขใƒ‡ใƒซใŒ่€ƒๆ…ฎใ™ใ‚‹ๅ˜่ชžใฎ็ฏ„ๅ›ฒใ‚’็‹ญใ‚ใ€ใ‚ˆใ‚Šไธ€่ฒซๆ€งใฎใ‚ใ‚‹ใƒ†ใ‚ญใ‚นใƒˆใ‚’็”Ÿๆˆใ™ใ‚‹ใ‚ˆใ†ใซใชใ‚Šใพใ™ใ€‚
32
  - ็”Ÿๆˆใ™ใ‚‹ๅ˜่ชžๆ•ฐ๏ผˆ-n๏ผ‰: ใ“ใฎๅ€คใ‚’ๆธ›ใ‚‰ใ™ใ“ใจใงใ€ใƒขใƒ‡ใƒซใŒ็”Ÿๆˆใ™ใ‚‹ใƒ†ใ‚ญใ‚นใƒˆใฎ้•ทใ•ใ‚’็Ÿญใใ—ใ€ไธ่ฆใช่ฟฝๅŠ ใƒ†ใ‚ญใ‚นใƒˆใฎ็”Ÿๆˆใ‚’้˜ฒใใ“ใจใŒใงใใพใ™ใ€‚-1 = ็„ก้™ๅคงใ€-2 = ๆ–‡่„ˆใŒๆบ€ใŸใ•ใ‚Œใ‚‹ใพใงใ€‚
33
 
34
+ ไปฅไธ‹ใฏllama.cppใฎไฝœ่€…(ggerganov)ใซใ‚ˆใ‚‹[ๆŽจๅฅจใƒ‘ใƒฉใƒกใƒผใ‚ฟใƒผ](https://huggingface.co/google/gemma-7b-it/discussions/38#65d7b14adb51f7c160769fa1)ใงใ™
35
  - -e (ๆ”น่กŒ\nใ‚’ใ‚จใ‚นใ‚ฑใƒผใƒ—)
36
  - --temp 0 (ๆœ€ใ‚‚็ขบ็Ž‡ใฎ้ซ˜ใ„ใƒˆใƒผใ‚ฏใƒณใฎใฟใ‚’้ธๆŠž)
37
  - --repeat-penalty 1.0 (็นฐใ‚Š่ฟ”ใ—ใƒšใƒŠใƒซใƒ†ใ‚ฃใ‚’ใ‚ชใƒ•ใ€‚ๆŒ‡็คบ่ชฟๆ•ดๆธˆใƒขใƒ‡ใƒซใงใ“ใ‚Œใ‚’ใ™ใ‚‹ใฎใฏใ€ๆฑบใ—ใฆ่‰ฏใ„่€ƒใˆใจใฏ่จ€ใˆใชใ„ใ€‚)
 
42
  - Top P (--top_p): Setting this value even lower will narrow the range of words considered by the model and produce more consistent text.
43
  - Number of words to generate (-n): Reducing this value will shorten the length of text generated by the model and prevent the generation of unnecessary additional text. -1 = infinity(default), -2 = until context filled.
44
 
45
+ The following are the [recommended parameters](https://huggingface.co/google/gemma-7b-it/discussions/38#65d7b14adb51f7c160769fa1) by the author of llama.cpp(ggerganov)
46
  - -e (escape newlines (\n))
47
  - --temp 0(pick most probable tokens)
48
  - --repeat-penalty 1.0(disable repetition penalty (it's never a good idea to have this with instruction tuned models)