mradermacher commited on
Commit
7f5c6cc
1 Parent(s): ed8f804

auto-patch README.md

Browse files
Files changed (1) hide show
  1. README.md +80 -0
README.md CHANGED
@@ -1,6 +1,86 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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/ContextualAI/archangel_sft-kto_llama30b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: ContextualAI/archangel_sft-kto_llama30b
3
+ datasets:
4
+ - stanfordnlp/SHP
5
+ - Anthropic/hh-rlhf
6
+ - OpenAssistant/oasst1
7
+ language:
8
+ - en
9
+ library_name: transformers
10
+ license: apache-2.0
11
+ quantized_by: mradermacher
12
+ tags:
13
+ - human feedback
14
+ - rlhf
15
+ - preferences
16
+ - alignment
17
+ - HALO
18
+ - halos
19
+ - dpo
20
+ - rl
21
+ ---
22
+ ## About
23
+
24
  <!-- ### quantize_version: 2 -->
25
  <!-- ### output_tensor_quantised: 1 -->
26
  <!-- ### convert_type: hf -->
27
  <!-- ### vocab_type: -->
28
  <!-- ### tags: nicoboss -->
29
  weighted/imatrix quants of https://huggingface.co/ContextualAI/archangel_sft-kto_llama30b
30
+
31
+ <!-- provided-files -->
32
+ static quants are available at https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-GGUF
33
+ ## Usage
34
+
35
+ If you are unsure how to use GGUF files, refer to one of [TheBloke's
36
+ READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
37
+ more details, including on how to concatenate multi-part files.
38
+
39
+ ## Provided Quants
40
+
41
+ (sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
42
+
43
+ | Link | Type | Size/GB | Notes |
44
+ |:-----|:-----|--------:|:------|
45
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-IQ1_S.gguf) | i1-IQ1_S | 7.2 | for the desperate |
46
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-IQ1_M.gguf) | i1-IQ1_M | 7.8 | mostly desperate |
47
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-IQ2_XXS.gguf) | i1-IQ2_XXS | 8.8 | |
48
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-IQ2_XS.gguf) | i1-IQ2_XS | 9.7 | |
49
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-IQ2_S.gguf) | i1-IQ2_S | 10.5 | |
50
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-IQ2_M.gguf) | i1-IQ2_M | 11.3 | |
51
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-Q2_K.gguf) | i1-Q2_K | 12.1 | IQ3_XXS probably better |
52
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 12.4 | lower quality |
53
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-IQ3_XS.gguf) | i1-IQ3_XS | 13.4 | |
54
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-IQ3_S.gguf) | i1-IQ3_S | 14.2 | beats Q3_K* |
55
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-Q3_K_S.gguf) | i1-Q3_K_S | 14.2 | IQ3_XS probably better |
56
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-IQ3_M.gguf) | i1-IQ3_M | 15.0 | |
57
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-Q3_K_M.gguf) | i1-Q3_K_M | 15.9 | IQ3_S probably better |
58
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-Q3_K_L.gguf) | i1-Q3_K_L | 17.4 | IQ3_M probably better |
59
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-IQ4_XS.gguf) | i1-IQ4_XS | 17.4 | |
60
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-Q4_0.gguf) | i1-Q4_0 | 18.5 | fast, low quality |
61
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-Q4_K_S.gguf) | i1-Q4_K_S | 18.6 | optimal size/speed/quality |
62
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-Q4_K_M.gguf) | i1-Q4_K_M | 19.7 | fast, recommended |
63
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-Q5_K_S.gguf) | i1-Q5_K_S | 22.5 | |
64
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-Q5_K_M.gguf) | i1-Q5_K_M | 23.1 | |
65
+ | [GGUF](https://huggingface.co/mradermacher/archangel_sft-kto_llama30b-i1-GGUF/resolve/main/archangel_sft-kto_llama30b.i1-Q6_K.gguf) | i1-Q6_K | 26.8 | practically like static Q6_K |
66
+
67
+ Here is a handy graph by ikawrakow comparing some lower-quality quant
68
+ types (lower is better):
69
+
70
+ ![image.png](https://www.nethype.de/huggingface_embed/quantpplgraph.png)
71
+
72
+ And here are Artefact2's thoughts on the matter:
73
+ https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
74
+
75
+ ## FAQ / Model Request
76
+
77
+ See https://huggingface.co/mradermacher/model_requests for some answers to
78
+ questions you might have and/or if you want some other model quantized.
79
+
80
+ ## Thanks
81
+
82
+ I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
83
+ me use its servers and providing upgrades to my workstation to enable
84
+ 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.
85
+
86
+ <!-- end -->