bhenrym14 commited on
Commit
a6e30d7
1 Parent(s): 6d2b533

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +9 -9
README.md CHANGED
@@ -5,7 +5,7 @@ datasets:
5
 
6
  Mostly untested!
7
 
8
- # RoPE Scaled QLoRA Fine-tune of Llama-13b on airoboros-gpt4-1.4.1 (GPTQ)
9
 
10
  ## Overview
11
 
@@ -61,15 +61,14 @@ Quantized with AutoGPTQ (bits = 4, group_size = 128, desc_act = True).
61
 
62
  See original model card below.
63
 
 
64
 
65
 
66
- # Original model card: Jon Durbin's Airoboros 7B GPT4 1.4
67
-
68
- __mostly untested, use if you want, or wait for some validation__
69
 
70
  ## Overview
71
 
72
- This is a __full__ (not qlora) fine-tune 7b parameter LlaMa model, using completely synthetic training data created gpt4 via https://github.com/jondurbin/airoboros
73
 
74
  This is mostly an extension of the previous gpt-4 series, with a few extras:
75
 
@@ -80,7 +79,7 @@ This is mostly an extension of the previous gpt-4 series, with a few extras:
80
  * riddles
81
  * all coding instructions have an equivalent " PLAINFORMAT" version now (and all rosettacode examples were trained with PLAINFORMAT)
82
 
83
- This model was fine-tuned with a fork of [FastChat](https://github.com/jondurbin/FastChat)
84
 
85
  The prompt it was trained with was:
86
 
@@ -103,7 +102,7 @@ Be sure you are pulling the latest branch!
103
  Then, you can invoke it like so (after downloading the model):
104
  ```
105
  python -m fastchat.serve.cli \
106
- --model-path airoboros-7b-gpt4-1.4 \
107
  --temperature 0.5 \
108
  --max-new-tokens 2048 \
109
  --no-history
@@ -203,10 +202,11 @@ Or:
203
  Write a multi-threaded TCP server in C that accepts a "GET [key]" input and "SET [key] [value]" input, and uses a binary tree to get and store the input values.
204
  ```
205
 
206
- You can optionally add a single space and "PLAINFORMAT" at the end of your prompt to avoid backticks, explanations, etc. and just print the code, e.g.:
207
 
208
  ```
209
- Write a websocket application in node.js. PLAINFORMAT
 
210
  ```
211
 
212
  ### Word games / trivia
 
5
 
6
  Mostly untested!
7
 
8
+ # RoPE Scaled QLoRA Fine-tune of Llama-33b on airoboros-gpt4-1.4.1 (GPTQ)
9
 
10
  ## Overview
11
 
 
61
 
62
  See original model card below.
63
 
64
+ # Original model card: Jon Durbin's Airoboros 33B GPT4 1.4
65
 
66
 
67
+ __not yet tested!__
 
 
68
 
69
  ## Overview
70
 
71
+ This is a qlora fine-tune 33b parameter LlaMa model, using completely synthetic training data created gpt4 via https://github.com/jondurbin/airoboros
72
 
73
  This is mostly an extension of the previous gpt-4 series, with a few extras:
74
 
 
79
  * riddles
80
  * all coding instructions have an equivalent " PLAINFORMAT" version now (and all rosettacode examples were trained with PLAINFORMAT)
81
 
82
+ This model was fine-tuned with a fork of [qlora](https://github.com/jondurbin/qlora)
83
 
84
  The prompt it was trained with was:
85
 
 
102
  Then, you can invoke it like so (after downloading the model):
103
  ```
104
  python -m fastchat.serve.cli \
105
+ --model-path airoboros-33b-gpt4-1.4 \
106
  --temperature 0.5 \
107
  --max-new-tokens 2048 \
108
  --no-history
 
202
  Write a multi-threaded TCP server in C that accepts a "GET [key]" input and "SET [key] [value]" input, and uses a binary tree to get and store the input values.
203
  ```
204
 
205
+ You can optionally add a newline and "PLAINFORMAT" at the end of your prompt to avoid backticks, explanations, etc. and just print the code, e.g.:
206
 
207
  ```
208
+ Write a websocket application in node.js.
209
+ PLAINFORMAT
210
  ```
211
 
212
  ### Word games / trivia