mradermacher commited on
Commit
ff98197
1 Parent(s): 1799304

auto-patch README.md

Browse files
Files changed (1) hide show
  1. README.md +69 -0
README.md CHANGED
@@ -1,5 +1,74 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
  <!-- ### quantize_version: 2 -->
2
  <!-- ### output_tensor_quantised: 1 -->
3
  <!-- ### convert_type: hf -->
4
  <!-- ### vocab_type: -->
5
  weighted/imatrix quants of https://huggingface.co/failspy/llama-3-70B-Instruct-abliterated
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: failspy/llama-3-70B-Instruct-abliterated
3
+ language:
4
+ - en
5
+ library_name: transformers
6
+ license: llama3
7
+ license_link: LICENSE
8
+ license_name: llama3
9
+ quantized_by: mradermacher
10
+ ---
11
+ ## About
12
+
13
  <!-- ### quantize_version: 2 -->
14
  <!-- ### output_tensor_quantised: 1 -->
15
  <!-- ### convert_type: hf -->
16
  <!-- ### vocab_type: -->
17
  weighted/imatrix quants of https://huggingface.co/failspy/llama-3-70B-Instruct-abliterated
18
+
19
+ <!-- provided-files -->
20
+ static quants are available at https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-GGUF
21
+ ## Usage
22
+
23
+ If you are unsure how to use GGUF files, refer to one of [TheBloke's
24
+ READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
25
+ more details, including on how to concatenate multi-part files.
26
+
27
+ ## Provided Quants
28
+
29
+ (sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
30
+
31
+ | Link | Type | Size/GB | Notes |
32
+ |:-----|:-----|--------:|:------|
33
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-IQ1_S.gguf) | i1-IQ1_S | 15.4 | for the desperate |
34
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-IQ1_M.gguf) | i1-IQ1_M | 16.9 | mostly desperate |
35
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-IQ2_XXS.gguf) | i1-IQ2_XXS | 19.2 | |
36
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-IQ2_XS.gguf) | i1-IQ2_XS | 21.2 | |
37
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-IQ2_S.gguf) | i1-IQ2_S | 22.3 | |
38
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-IQ2_M.gguf) | i1-IQ2_M | 24.2 | |
39
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-Q2_K.gguf) | i1-Q2_K | 26.5 | IQ3_XXS probably better |
40
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 27.6 | lower quality |
41
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-IQ3_XS.gguf) | i1-IQ3_XS | 29.4 | |
42
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-IQ3_S.gguf) | i1-IQ3_S | 31.0 | beats Q3_K* |
43
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-Q3_K_S.gguf) | i1-Q3_K_S | 31.0 | IQ3_XS probably better |
44
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-IQ3_M.gguf) | i1-IQ3_M | 32.0 | |
45
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-Q3_K_M.gguf) | i1-Q3_K_M | 34.4 | IQ3_S probably better |
46
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-Q3_K_L.gguf) | i1-Q3_K_L | 37.2 | IQ3_M probably better |
47
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-IQ4_XS.gguf) | i1-IQ4_XS | 38.0 | |
48
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-Q4_0.gguf) | i1-Q4_0 | 40.2 | fast, low quality |
49
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-Q4_K_S.gguf) | i1-Q4_K_S | 40.4 | optimal size/speed/quality |
50
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-Q4_K_M.gguf) | i1-Q4_K_M | 42.6 | fast, recommended |
51
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-Q5_K_S.gguf) | i1-Q5_K_S | 48.8 | |
52
+ | [GGUF](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-Q5_K_M.gguf) | i1-Q5_K_M | 50.0 | |
53
+ | [PART 1](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-Q6_K.gguf.part1of2) [PART 2](https://huggingface.co/mradermacher/llama-3-70B-Instruct-abliterated-i1-GGUF/resolve/main/llama-3-70B-Instruct-abliterated.i1-Q6_K.gguf.part2of2) | i1-Q6_K | 58.0 | practically like static Q6_K |
54
+
55
+ Here is a handy graph by ikawrakow comparing some lower-quality quant
56
+ types (lower is better):
57
+
58
+ ![image.png](https://www.nethype.de/huggingface_embed/quantpplgraph.png)
59
+
60
+ And here are Artefact2's thoughts on the matter:
61
+ https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
62
+
63
+ ## FAQ / Model Request
64
+
65
+ See https://huggingface.co/mradermacher/model_requests for some answers to
66
+ questions you might have and/or if you want some other model quantized.
67
+
68
+ ## Thanks
69
+
70
+ I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
71
+ me use its servers and providing upgrades to my workstation to enable
72
+ this work in my free time.
73
+
74
+ <!-- end -->