mradermacher commited on
Commit
0cdc134
1 Parent(s): d9b6e13

auto-patch README.md

Browse files
Files changed (1) hide show
  1. README.md +20 -0
README.md CHANGED
@@ -71,10 +71,30 @@ more details, including on how to concatenate multi-part files.
71
 
72
  | Link | Type | Size/GB | Notes |
73
  |:-----|:-----|--------:|:------|
 
 
 
 
 
 
74
  | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-Q2_K.gguf) | i1-Q2_K | 2.8 | IQ3_XXS probably better |
75
  | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 2.9 | lower quality |
 
 
 
76
  | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-IQ3_M.gguf) | i1-IQ3_M | 3.4 | |
 
 
 
 
 
 
 
77
  | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-Q4_K_S.gguf) | i1-Q4_K_S | 4.2 | optimal size/speed/quality |
 
 
 
 
78
 
79
  Here is a handy graph by ikawrakow comparing some lower-quality quant
80
  types (lower is better):
 
71
 
72
  | Link | Type | Size/GB | Notes |
73
  |:-----|:-----|--------:|:------|
74
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-IQ1_S.gguf) | i1-IQ1_S | 1.7 | for the desperate |
75
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-IQ1_M.gguf) | i1-IQ1_M | 1.9 | mostly desperate |
76
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-IQ2_XXS.gguf) | i1-IQ2_XXS | 2.1 | |
77
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-IQ2_XS.gguf) | i1-IQ2_XS | 2.3 | |
78
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-IQ2_S.gguf) | i1-IQ2_S | 2.4 | |
79
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-IQ2_M.gguf) | i1-IQ2_M | 2.6 | |
80
  | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-Q2_K.gguf) | i1-Q2_K | 2.8 | IQ3_XXS probably better |
81
  | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 2.9 | lower quality |
82
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-IQ3_XS.gguf) | i1-IQ3_XS | 3.1 | |
83
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-Q3_K_S.gguf) | i1-Q3_K_S | 3.3 | IQ3_XS probably better |
84
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-IQ3_S.gguf) | i1-IQ3_S | 3.3 | beats Q3_K* |
85
  | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-IQ3_M.gguf) | i1-IQ3_M | 3.4 | |
86
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-Q3_K_M.gguf) | i1-Q3_K_M | 3.6 | IQ3_S probably better |
87
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-Q3_K_L.gguf) | i1-Q3_K_L | 3.9 | IQ3_M probably better |
88
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-IQ4_XS.gguf) | i1-IQ4_XS | 4.0 | |
89
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-Q4_0_4_4.gguf) | i1-Q4_0_4_4 | 4.2 | fast on arm, low quality |
90
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-Q4_0_4_8.gguf) | i1-Q4_0_4_8 | 4.2 | fast on arm+i8mm, low quality |
91
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-Q4_0_8_8.gguf) | i1-Q4_0_8_8 | 4.2 | fast on arm+sve, low quality |
92
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-Q4_0.gguf) | i1-Q4_0 | 4.2 | fast, low quality |
93
  | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-Q4_K_S.gguf) | i1-Q4_K_S | 4.2 | optimal size/speed/quality |
94
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-Q4_K_M.gguf) | i1-Q4_K_M | 4.5 | fast, recommended |
95
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-Q5_K_S.gguf) | i1-Q5_K_S | 5.1 | |
96
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-Q5_K_M.gguf) | i1-Q5_K_M | 5.2 | |
97
+ | [GGUF](https://huggingface.co/mradermacher/bagel-dpo-7b-v0.4-i1-GGUF/resolve/main/bagel-dpo-7b-v0.4.i1-Q6_K.gguf) | i1-Q6_K | 6.0 | practically like static Q6_K |
98
 
99
  Here is a handy graph by ikawrakow comparing some lower-quality quant
100
  types (lower is better):