Hugging Face's logo

bartowski
/
Mistral7B-PairRM-SPPO-Iter3-GGUF

Text Generation
GGUF
English
Model card Files Community
bartowski commited on
Commit
cdee304
•
1 Parent(s): f1d930e

Llamacpp quants

Browse files
Files changed (23) hide show
  1. .gitattributes +21 -0
  2. Mistral7B-PairRM-SPPO-Iter3-IQ2_M.gguf +3 -0
  3. Mistral7B-PairRM-SPPO-Iter3-IQ2_S.gguf +3 -0
  4. Mistral7B-PairRM-SPPO-Iter3-IQ2_XS.gguf +3 -0
  5. Mistral7B-PairRM-SPPO-Iter3-IQ3_M.gguf +3 -0
  6. Mistral7B-PairRM-SPPO-Iter3-IQ3_S.gguf +3 -0
  7. Mistral7B-PairRM-SPPO-Iter3-IQ3_XS.gguf +3 -0
  8. Mistral7B-PairRM-SPPO-Iter3-IQ3_XXS.gguf +3 -0
  9. Mistral7B-PairRM-SPPO-Iter3-IQ4_NL.gguf +3 -0
  10. Mistral7B-PairRM-SPPO-Iter3-IQ4_XS.gguf +3 -0
  11. Mistral7B-PairRM-SPPO-Iter3-Q2_K.gguf +3 -0
  12. Mistral7B-PairRM-SPPO-Iter3-Q3_K_L.gguf +3 -0
  13. Mistral7B-PairRM-SPPO-Iter3-Q3_K_M.gguf +3 -0
  14. Mistral7B-PairRM-SPPO-Iter3-Q3_K_S.gguf +3 -0
  15. Mistral7B-PairRM-SPPO-Iter3-Q4_K_M.gguf +3 -0
  16. Mistral7B-PairRM-SPPO-Iter3-Q4_K_S.gguf +3 -0
  17. Mistral7B-PairRM-SPPO-Iter3-Q5_K_M.gguf +3 -0
  18. Mistral7B-PairRM-SPPO-Iter3-Q5_K_S.gguf +3 -0
  19. Mistral7B-PairRM-SPPO-Iter3-Q6_K.gguf +3 -0
  20. Mistral7B-PairRM-SPPO-Iter3-Q8_0.gguf +3 -0
  21. Mistral7B-PairRM-SPPO-Iter3-fp16.gguf +3 -0
  22. Mistral7B-PairRM-SPPO-Iter3.imatrix +3 -0
  23. README.md +100 -0
.gitattributes CHANGED
@@ -33,3 +33,24 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ Mistral7B-PairRM-SPPO-Iter3-IQ2_M.gguf filter=lfs diff=lfs merge=lfs -text
37
+ Mistral7B-PairRM-SPPO-Iter3-IQ2_S.gguf filter=lfs diff=lfs merge=lfs -text
38
+ Mistral7B-PairRM-SPPO-Iter3-IQ2_XS.gguf filter=lfs diff=lfs merge=lfs -text
39
+ Mistral7B-PairRM-SPPO-Iter3-IQ3_M.gguf filter=lfs diff=lfs merge=lfs -text
40
+ Mistral7B-PairRM-SPPO-Iter3-IQ3_S.gguf filter=lfs diff=lfs merge=lfs -text
41
+ Mistral7B-PairRM-SPPO-Iter3-IQ3_XS.gguf filter=lfs diff=lfs merge=lfs -text
42
+ Mistral7B-PairRM-SPPO-Iter3-IQ3_XXS.gguf filter=lfs diff=lfs merge=lfs -text
43
+ Mistral7B-PairRM-SPPO-Iter3-IQ4_NL.gguf filter=lfs diff=lfs merge=lfs -text
44
+ Mistral7B-PairRM-SPPO-Iter3-IQ4_XS.gguf filter=lfs diff=lfs merge=lfs -text
45
+ Mistral7B-PairRM-SPPO-Iter3-Q2_K.gguf filter=lfs diff=lfs merge=lfs -text
46
+ Mistral7B-PairRM-SPPO-Iter3-Q3_K_L.gguf filter=lfs diff=lfs merge=lfs -text
47
+ Mistral7B-PairRM-SPPO-Iter3-Q3_K_M.gguf filter=lfs diff=lfs merge=lfs -text
48
+ Mistral7B-PairRM-SPPO-Iter3-Q3_K_S.gguf filter=lfs diff=lfs merge=lfs -text
49
+ Mistral7B-PairRM-SPPO-Iter3-Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text
50
+ Mistral7B-PairRM-SPPO-Iter3-Q4_K_S.gguf filter=lfs diff=lfs merge=lfs -text
51
+ Mistral7B-PairRM-SPPO-Iter3-Q5_K_M.gguf filter=lfs diff=lfs merge=lfs -text
52
+ Mistral7B-PairRM-SPPO-Iter3-Q5_K_S.gguf filter=lfs diff=lfs merge=lfs -text
53
+ Mistral7B-PairRM-SPPO-Iter3-Q6_K.gguf filter=lfs diff=lfs merge=lfs -text
54
+ Mistral7B-PairRM-SPPO-Iter3-Q8_0.gguf filter=lfs diff=lfs merge=lfs -text
55
+ Mistral7B-PairRM-SPPO-Iter3-fp16.gguf filter=lfs diff=lfs merge=lfs -text
56
+ Mistral7B-PairRM-SPPO-Iter3.imatrix filter=lfs diff=lfs merge=lfs -text
Mistral7B-PairRM-SPPO-Iter3-IQ2_M.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:087428dc5844f03ebe3d96b993bb0e71d056d37e984fb5eb8162d6bb6da461fe
3
+ size 2500713088
Mistral7B-PairRM-SPPO-Iter3-IQ2_S.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a1a1c205d5221d3d3372d16d911d91caea4fc8b4c2e1504109739498eefea8bd
3
+ size 2310920832
Mistral7B-PairRM-SPPO-Iter3-IQ2_XS.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7a1cebb88e2ff14a56b1cf381079d9871583cb0efa200a2b84106fd584b01096
3
+ size 2198256256
Mistral7B-PairRM-SPPO-Iter3-IQ3_M.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:73d5951fdce872b0539e1dc0af16e3b83c87c11aecc607d4c863c5e7287a580a
3
+ size 3284892288
Mistral7B-PairRM-SPPO-Iter3-IQ3_S.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b4a18f826c0bbfdfe3014134142684e833cd084c083231f7b37324c077180aa0
3
+ size 3182393984
Mistral7B-PairRM-SPPO-Iter3-IQ3_XS.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4ceae69545c8dbf9282f2069f0167e67b1ebea7e634e89063331f6786eb1e187
3
+ size 3018816128
Mistral7B-PairRM-SPPO-Iter3-IQ3_XXS.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d0b3aeb21f6a31f173a1398456e85f365eb36970e7c17c24fcc6efc5e746e9f
3
+ size 2827344512
Mistral7B-PairRM-SPPO-Iter3-IQ4_NL.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d60650b5431f57130c0732f95a1bd10af7baedf8bb1637facad8b36b376b08da
3
+ size 4125694592
Mistral7B-PairRM-SPPO-Iter3-IQ4_XS.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5fc429bc60e3f4112e803818d00777d3b1516c0f6bfccdb9d48a408243edfe12
3
+ size 3907689088
Mistral7B-PairRM-SPPO-Iter3-Q2_K.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cb60c5527a87e2ed29dd64ff11ddaa0ac7a84370aca3161f402c3b8bb1a29f87
3
+ size 2719242880
Mistral7B-PairRM-SPPO-Iter3-Q3_K_L.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ea10064036a4f802fb53e293f65445d70f716db2c5f4de6636b438fd53f5340c
3
+ size 3822025344
Mistral7B-PairRM-SPPO-Iter3-Q3_K_M.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e0f1af44bbaeb5e78a43c26e4b84c6875443a5e23986d9820af1cdc3dcb843d7
3
+ size 3518986880
Mistral7B-PairRM-SPPO-Iter3-Q3_K_S.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8e284c0f992a0832c90321246ef4bd2b5e37d646f117a1294a9350be0e7c4831
3
+ size 3164568192
Mistral7B-PairRM-SPPO-Iter3-Q4_K_M.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0c2cfb2f98778f53c57a590a18443396b10371732c6366e3cba8d74acd33b620
3
+ size 4368439936
Mistral7B-PairRM-SPPO-Iter3-Q4_K_S.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:33e7b754fb2960513fa912a3484ffcce49f2796e030d596846b45701976ddaf2
3
+ size 4140374656
Mistral7B-PairRM-SPPO-Iter3-Q5_K_M.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7b2483761dd3aa6699f980043f3e5de7fa78b924aeb85e87b65e9c048cb3f52a
3
+ size 5131410048
Mistral7B-PairRM-SPPO-Iter3-Q5_K_S.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0cb17d628edc97cd89a43cb7d1082e1a46f09bf35de251ef4ac1fd0355f7a91b
3
+ size 4997716608
Mistral7B-PairRM-SPPO-Iter3-Q6_K.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eae2e55c7cc702cc9ceb252ea2d29c45cedb3c0172ea551cdb4e8efc9b891405
3
+ size 5942065792
Mistral7B-PairRM-SPPO-Iter3-Q8_0.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:568edd441f3b7a53335a9dd8e2af1e1377a9bdb04d76d71ebd7e691763883659
3
+ size 7695858304
Mistral7B-PairRM-SPPO-Iter3-fp16.gguf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c30b6256ca712f02d8e79c2f7fba5a9123011f95a4e0a4f739b940b9842a4915
3
+ size 14484732192
Mistral7B-PairRM-SPPO-Iter3.imatrix ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:849a401647d59d6815167ddb73bc845376f40f0f2205fe591eaa48cee4c58cb0
3
+ size 4988166
README.md ADDED
@@ -0,0 +1,100 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ datasets:
4
+ - openbmb/UltraFeedback
5
+ language:
6
+ - en
7
+ pipeline_tag: text-generation
8
+ quantized_by: bartowski
9
+ ---
10
+
11
+ ## Llamacpp imatrix Quantizations of Mistral7B-PairRM-SPPO-Iter3
12
+
13
+ Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b2777">b2777</a> for quantization.
14
+
15
+ Original model: https://huggingface.co/UCLA-AGI/Mistral7B-PairRM-SPPO-Iter3
16
+
17
+ All quants made using imatrix option with dataset provided by Kalomaze [here](https://github.com/ggerganov/llama.cpp/discussions/5263#discussioncomment-8395384)
18
+
19
+ ## Prompt format
20
+
21
+ ```
22
+ <s> [INST] {prompt} [/INST]</s>
23
+ ```
24
+
25
+ Note that this model does not support a System prompt.
26
+
27
+ ## Download a file (not the whole branch) from below:
28
+
29
+ | Filename | Quant type | File Size | Description |
30
+ | -------- | ---------- | --------- | ----------- |
31
+ | [Mistral7B-PairRM-SPPO-Iter3-Q8_0.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-Q8_0.gguf) | Q8_0 | 7.69GB | Extremely high quality, generally unneeded but max available quant. |
32
+ | [Mistral7B-PairRM-SPPO-Iter3-Q6_K.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-Q6_K.gguf) | Q6_K | 5.94GB | Very high quality, near perfect, *recommended*. |
33
+ | [Mistral7B-PairRM-SPPO-Iter3-Q5_K_M.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-Q5_K_M.gguf) | Q5_K_M | 5.13GB | High quality, *recommended*. |
34
+ | [Mistral7B-PairRM-SPPO-Iter3-Q5_K_S.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-Q5_K_S.gguf) | Q5_K_S | 4.99GB | High quality, *recommended*. |
35
+ | [Mistral7B-PairRM-SPPO-Iter3-Q4_K_M.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-Q4_K_M.gguf) | Q4_K_M | 4.36GB | Good quality, uses about 4.83 bits per weight, *recommended*. |
36
+ | [Mistral7B-PairRM-SPPO-Iter3-Q4_K_S.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-Q4_K_S.gguf) | Q4_K_S | 4.14GB | Slightly lower quality with more space savings, *recommended*. |
37
+ | [Mistral7B-PairRM-SPPO-Iter3-IQ4_NL.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-IQ4_NL.gguf) | IQ4_NL | 4.12GB | Decent quality, slightly smaller than Q4_K_S with similar performance *recommended*. |
38
+ | [Mistral7B-PairRM-SPPO-Iter3-IQ4_XS.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-IQ4_XS.gguf) | IQ4_XS | 3.90GB | Decent quality, smaller than Q4_K_S with similar performance, *recommended*. |
39
+ | [Mistral7B-PairRM-SPPO-Iter3-Q3_K_L.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-Q3_K_L.gguf) | Q3_K_L | 3.82GB | Lower quality but usable, good for low RAM availability. |
40
+ | [Mistral7B-PairRM-SPPO-Iter3-Q3_K_M.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-Q3_K_M.gguf) | Q3_K_M | 3.51GB | Even lower quality. |
41
+ | [Mistral7B-PairRM-SPPO-Iter3-IQ3_M.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-IQ3_M.gguf) | IQ3_M | 3.28GB | Medium-low quality, new method with decent performance comparable to Q3_K_M. |
42
+ | [Mistral7B-PairRM-SPPO-Iter3-IQ3_S.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-IQ3_S.gguf) | IQ3_S | 3.18GB | Lower quality, new method with decent performance, recommended over Q3_K_S quant, same size with better performance. |
43
+ | [Mistral7B-PairRM-SPPO-Iter3-Q3_K_S.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-Q3_K_S.gguf) | Q3_K_S | 3.16GB | Low quality, not recommended. |
44
+ | [Mistral7B-PairRM-SPPO-Iter3-IQ3_XS.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-IQ3_XS.gguf) | IQ3_XS | 3.01GB | Lower quality, new method with decent performance, slightly better than Q3_K_S. |
45
+ | [Mistral7B-PairRM-SPPO-Iter3-IQ3_XXS.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-IQ3_XXS.gguf) | IQ3_XXS | 2.82GB | Lower quality, new method with decent performance, comparable to Q3 quants. |
46
+ | [Mistral7B-PairRM-SPPO-Iter3-Q2_K.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-Q2_K.gguf) | Q2_K | 2.71GB | Very low quality but surprisingly usable. |
47
+ | [Mistral7B-PairRM-SPPO-Iter3-IQ2_M.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-IQ2_M.gguf) | IQ2_M | 2.50GB | Very low quality, uses SOTA techniques to also be surprisingly usable. |
48
+ | [Mistral7B-PairRM-SPPO-Iter3-IQ2_S.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-IQ2_S.gguf) | IQ2_S | 2.31GB | Very low quality, uses SOTA techniques to be usable. |
49
+ | [Mistral7B-PairRM-SPPO-Iter3-IQ2_XS.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF/blob/main/Mistral7B-PairRM-SPPO-Iter3-IQ2_XS.gguf) | IQ2_XS | 2.19GB | Very low quality, uses SOTA techniques to be usable. |
50
+ | [Mistral7B-PairRM-SPPO-Iter3-IQ2_XXS.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF//main/Mistral7B-PairRM-SPPO-Iter3-IQ2_XXS.gguf) | IQ2_XXS | | Lower quality, uses SOTA techniques to be usable. |
51
+ | [Mistral7B-PairRM-SPPO-Iter3-IQ1_M.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF//main/Mistral7B-PairRM-SPPO-Iter3-IQ1_M.gguf) | IQ1_M | | Extremely low quality, *not* recommended. |
52
+ | [Mistral7B-PairRM-SPPO-Iter3-IQ1_S.gguf](https://huggingface.co/bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF//main/Mistral7B-PairRM-SPPO-Iter3-IQ1_S.gguf) | IQ1_S | | Extremely low quality, *not* recommended. |
53
+
54
+ ## Downloading using huggingface-cli
55
+
56
+ First, make sure you have hugginface-cli installed:
57
+
58
+ ```
59
+ pip install -U "huggingface_hub[cli]"
60
+ ```
61
+
62
+ Then, you can target the specific file you want:
63
+
64
+ ```
65
+ huggingface-cli download bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF --include "Mistral7B-PairRM-SPPO-Iter3-Q4_K_M.gguf" --local-dir ./ --local-dir-use-symlinks False
66
+ ```
67
+
68
+ If the model is bigger than 50GB, it will have been split into multiple files. In order to download them all to a local folder, run:
69
+
70
+ ```
71
+ huggingface-cli download bartowski/Mistral7B-PairRM-SPPO-Iter3-GGUF --include "Mistral7B-PairRM-SPPO-Iter3-Q8_0.gguf/*" --local-dir Mistral7B-PairRM-SPPO-Iter3-Q8_0 --local-dir-use-symlinks False
72
+ ```
73
+
74
+ You can either specify a new local-dir (Mistral7B-PairRM-SPPO-Iter3-Q8_0) or download them all in place (./)
75
+
76
+ ## Which file should I choose?
77
+
78
+ A great write up with charts showing various performances is provided by Artefact2 [here](https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9)
79
+
80
+ The first thing to figure out is how big a model you can run. To do this, you'll need to figure out how much RAM and/or VRAM you have.
81
+
82
+ If you want your model running as FAST as possible, you'll want to fit the whole thing on your GPU's VRAM. Aim for a quant with a file size 1-2GB smaller than your GPU's total VRAM.
83
+
84
+ If you want the absolute maximum quality, add both your system RAM and your GPU's VRAM together, then similarly grab a quant with a file size 1-2GB Smaller than that total.
85
+
86
+ Next, you'll need to decide if you want to use an 'I-quant' or a 'K-quant'.
87
+
88
+ If you don't want to think too much, grab one of the K-quants. These are in format 'QX_K_X', like Q5_K_M.
89
+
90
+ If you want to get more into the weeds, you can check out this extremely useful feature chart:
91
+
92
+ [llama.cpp feature matrix](https://github.com/ggerganov/llama.cpp/wiki/Feature-matrix)
93
+
94
+ But basically, if you're aiming for below Q4, and you're running cuBLAS (Nvidia) or rocBLAS (AMD), you should look towards the I-quants. These are in format IQX_X, like IQ3_M. These are newer and offer better performance for their size.
95
+
96
+ These I-quants can also be used on CPU and Apple Metal, but will be slower than their K-quant equivalent, so speed vs performance is a tradeoff you'll have to decide.
97
+
98
+ The I-quants are *not* compatible with Vulcan, which is also AMD, so if you have an AMD card double check if you're using the rocBLAS build or the Vulcan build. At the time of writing this, LM Studio has a preview with ROCm support, and other inference engines have specific builds for ROCm.
99
+
100
+ Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski