mradermacher commited on
Commit
2996e80
1 Parent(s): 4336adc

auto-patch README.md

Browse files
Files changed (1) hide show
  1. README.md +66 -0
README.md CHANGED
@@ -1,6 +1,72 @@
 
 
 
 
 
 
 
 
 
1
  <!-- ### quantize_version: 2 -->
2
  <!-- ### output_tensor_quantised: 1 -->
3
  <!-- ### convert_type: hf -->
4
  <!-- ### vocab_type: -->
5
  <!-- ### tags: nicoboss -->
6
  weighted/imatrix quants of https://huggingface.co/meta-llama/Meta-Llama-3.1-70B
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: meta-llama/Meta-Llama-3.1-70B
3
+ language:
4
+ - en
5
+ library_name: transformers
6
+ quantized_by: mradermacher
7
+ ---
8
+ ## About
9
+
10
  <!-- ### quantize_version: 2 -->
11
  <!-- ### output_tensor_quantised: 1 -->
12
  <!-- ### convert_type: hf -->
13
  <!-- ### vocab_type: -->
14
  <!-- ### tags: nicoboss -->
15
  weighted/imatrix quants of https://huggingface.co/meta-llama/Meta-Llama-3.1-70B
16
+
17
+ <!-- provided-files -->
18
+ static quants are available at https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-GGUF
19
+ ## Usage
20
+
21
+ If you are unsure how to use GGUF files, refer to one of [TheBloke's
22
+ READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
23
+ more details, including on how to concatenate multi-part files.
24
+
25
+ ## Provided Quants
26
+
27
+ (sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
28
+
29
+ | Link | Type | Size/GB | Notes |
30
+ |:-----|:-----|--------:|:------|
31
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-IQ1_S.gguf) | i1-IQ1_S | 15.4 | for the desperate |
32
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-IQ1_M.gguf) | i1-IQ1_M | 16.9 | mostly desperate |
33
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-IQ2_XXS.gguf) | i1-IQ2_XXS | 19.2 | |
34
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-IQ2_XS.gguf) | i1-IQ2_XS | 21.2 | |
35
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-IQ2_S.gguf) | i1-IQ2_S | 22.3 | |
36
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-IQ2_M.gguf) | i1-IQ2_M | 24.2 | |
37
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-Q2_K.gguf) | i1-Q2_K | 26.5 | IQ3_XXS probably better |
38
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 27.6 | lower quality |
39
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-IQ3_XS.gguf) | i1-IQ3_XS | 29.4 | |
40
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-IQ3_S.gguf) | i1-IQ3_S | 31.0 | beats Q3_K* |
41
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-Q3_K_S.gguf) | i1-Q3_K_S | 31.0 | IQ3_XS probably better |
42
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-IQ3_M.gguf) | i1-IQ3_M | 32.0 | |
43
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-Q3_K_M.gguf) | i1-Q3_K_M | 34.4 | IQ3_S probably better |
44
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-Q3_K_L.gguf) | i1-Q3_K_L | 37.2 | IQ3_M probably better |
45
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-IQ4_XS.gguf) | i1-IQ4_XS | 38.0 | |
46
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-Q4_0.gguf) | i1-Q4_0 | 40.2 | fast, low quality |
47
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-Q4_K_S.gguf) | i1-Q4_K_S | 40.4 | optimal size/speed/quality |
48
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-Q4_K_M.gguf) | i1-Q4_K_M | 42.6 | fast, recommended |
49
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-Q5_K_S.gguf) | i1-Q5_K_S | 48.8 | |
50
+ | [GGUF](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-Q5_K_M.gguf) | i1-Q5_K_M | 50.0 | |
51
+ | [PART 1](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-Q6_K.gguf.part1of2) [PART 2](https://huggingface.co/mradermacher/Meta-Llama-3.1-70B-i1-GGUF/resolve/main/Meta-Llama-3.1-70B.i1-Q6_K.gguf.part2of2) | i1-Q6_K | 58.0 | practically like static Q6_K |
52
+
53
+ Here is a handy graph by ikawrakow comparing some lower-quality quant
54
+ types (lower is better):
55
+
56
+ ![image.png](https://www.nethype.de/huggingface_embed/quantpplgraph.png)
57
+
58
+ And here are Artefact2's thoughts on the matter:
59
+ https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
60
+
61
+ ## FAQ / Model Request
62
+
63
+ See https://huggingface.co/mradermacher/model_requests for some answers to
64
+ questions you might have and/or if you want some other model quantized.
65
+
66
+ ## Thanks
67
+
68
+ I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
69
+ me use its servers and providing upgrades to my workstation to enable
70
+ this work in my free time. Additional thanks to [@nicoboss](https://huggingface.co/nicoboss) for giving me access to his hardware for calculating the imatrix for these quants.
71
+
72
+ <!-- end -->