mradermacher commited on
Commit
066239d
1 Parent(s): 589beff

auto-patch README.md

Browse files
Files changed (1) hide show
  1. README.md +59 -0
README.md CHANGED
@@ -1,5 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  <!-- ### quantize_version: 1 -->
2
  <!-- ### output_tensor_quantised: 1 -->
3
  <!-- ### convert_type: -->
4
  <!-- ### vocab_type: -->
5
  weighted/imatrix quants of https://huggingface.co/ibivibiv/hydra-moe-120b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ exported_from: ibivibiv/hydra-moe-120b
3
+ language:
4
+ - en
5
+ library_name: transformers
6
+ license: apache-2.0
7
+ quantized_by: mradermacher
8
+ tags:
9
+ - moe
10
+ - moerge
11
+ ---
12
+ ## About
13
+
14
  <!-- ### quantize_version: 1 -->
15
  <!-- ### output_tensor_quantised: 1 -->
16
  <!-- ### convert_type: -->
17
  <!-- ### vocab_type: -->
18
  weighted/imatrix quants of https://huggingface.co/ibivibiv/hydra-moe-120b
19
+
20
+
21
+ <!-- provided-files -->
22
+ static quants are available at https://huggingface.co/mradermacher/hydra-moe-120b-GGUF
23
+ ## Usage
24
+
25
+ If you are unsure how to use GGUF files, refer to one of [TheBloke's
26
+ READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
27
+ more details, including on how to concatenate multi-part files.
28
+
29
+ ## Provided Quants
30
+
31
+ (sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
32
+
33
+ | Link | Type | Size/GB | Notes |
34
+ |:-----|:-----|--------:|:------|
35
+ | [GGUF](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-IQ2_M.gguf) | i1-IQ2_M | 37.3 | |
36
+ | [GGUF](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q2_K.gguf) | i1-Q2_K | 41.6 | IQ3_XXS probably better |
37
+ | [GGUF](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 43.8 | lower quality |
38
+ | [GGUF](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q3_K_S.gguf) | i1-Q3_K_S | 49.1 | IQ3_XS probably better |
39
+ | [PART 1](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q3_K_M.gguf.part1of2) [PART 2](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q3_K_M.gguf.part2of2) | i1-Q3_K_M | 54.5 | IQ3_S probably better |
40
+ | [PART 1](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q3_K_L.gguf.part1of2) [PART 2](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q3_K_L.gguf.part2of2) | i1-Q3_K_L | 59.1 | IQ3_M probably better |
41
+ | [PART 1](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-IQ4_XS.gguf.part1of2) [PART 2](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-IQ4_XS.gguf.part2of2) | i1-IQ4_XS | 60.7 | |
42
+ | [PART 1](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q4_0.gguf.part1of2) [PART 2](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q4_0.gguf.part2of2) | i1-Q4_0 | 64.4 | fast, low quality |
43
+ | [PART 1](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q4_K_S.gguf.part1of2) [PART 2](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q4_K_S.gguf.part2of2) | i1-Q4_K_S | 64.7 | optimal size/speed/quality |
44
+ | [PART 1](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q4_K_M.gguf.part1of2) [PART 2](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q4_K_M.gguf.part2of2) | i1-Q4_K_M | 68.8 | fast, recommended |
45
+ | [PART 1](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q5_K_S.gguf.part1of2) [PART 2](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q5_K_S.gguf.part2of2) | i1-Q5_K_S | 78.3 | |
46
+ | [PART 1](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q5_K_M.gguf.part1of2) [PART 2](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q5_K_M.gguf.part2of2) | i1-Q5_K_M | 80.7 | |
47
+ | [PART 1](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q6_K.gguf.part1of2) [PART 2](https://huggingface.co/mradermacher/hydra-moe-120b-i1-GGUF/resolve/main/hydra-moe-120b.i1-Q6_K.gguf.part2of2) | i1-Q6_K | 93.3 | practically like static Q6_K |
48
+
49
+
50
+ Here is a handy graph by ikawrakow comparing some lower-quality quant
51
+ types (lower is better):
52
+
53
+ ![image.png](https://www.nethype.de/huggingface_embed/quantpplgraph.png)
54
+
55
+ And here are Artefact2's thoughts on the matter:
56
+ https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
57
+
58
+ ## Thanks
59
+
60
+ I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
61
+ me use its servers and providing upgrades to my workstation to enable
62
+ this work in my free time.
63
+
64
+ <!-- end -->