README.md
5.9 KB · 93 lines · markdown Raw
1 ---
2 base_model: servantofares/AReaL-SEA-235B-A22B
3 language:
4 - en
5 library_name: transformers
6 license: apache-2.0
7 mradermacher:
8 readme_rev: 1
9 quantized_by: mradermacher
10 tags:
11 - tool-use
12 - agent
13 - multi-turn
14 - reinforcement-learning
15 - GRPO
16 - tau2-bench
17 - AReaL
18 ---
19 ## About
20
21 <!-- ### quantize_version: 2 -->
22 <!-- ### output_tensor_quantised: 1 -->
23 <!-- ### convert_type: hf -->
24 <!-- ### vocab_type: -->
25 <!-- ### tags: nicoboss -->
26 <!-- ### quants: Q2_K IQ3_M Q4_K_S IQ3_XXS Q3_K_M small-IQ4_NL Q4_K_M IQ2_M Q6_K IQ4_XS Q2_K_S IQ1_M Q3_K_S IQ2_XXS Q3_K_L IQ2_XS Q5_K_S IQ2_S IQ1_S Q5_K_M Q4_0 IQ3_XS Q4_1 IQ3_S -->
27 <!-- ### quants_skip: -->
28 <!-- ### skip_mmproj: -->
29 weighted/imatrix quants of https://huggingface.co/servantofares/AReaL-SEA-235B-A22B
30
31 <!-- provided-files -->
32
33 ***For a convenient overview and download list, visit our [model page for this model](https://hf.tst.eu/model#AReaL-SEA-235B-A22B-i1-GGUF).***
34
35 static quants are available at https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-GGUF
36 ## Usage
37
38 If you are unsure how to use GGUF files, refer to one of [TheBloke's
39 READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
40 more details, including on how to concatenate multi-part files.
41
42 ## Provided Quants
43
44 (sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
45
46 | Link | Type | Size/GB | Notes |
47 |:-----|:-----|--------:|:------|
48 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.imatrix.gguf) | imatrix | 0.6 | imatrix file (for creating your own quants) |
49 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-IQ1_S.gguf) | i1-IQ1_S | 48.0 | for the desperate |
50 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-IQ1_M.gguf) | i1-IQ1_M | 53.2 | mostly desperate |
51 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-IQ2_XXS.gguf) | i1-IQ2_XXS | 61.9 | |
52 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-IQ2_XS.gguf) | i1-IQ2_XS | 68.9 | |
53 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-IQ2_S.gguf) | i1-IQ2_S | 70.3 | |
54 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-IQ2_M.gguf) | i1-IQ2_M | 77.3 | |
55 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-Q2_K_S.gguf) | i1-Q2_K_S | 79.9 | very low quality |
56 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-Q2_K.gguf) | i1-Q2_K | 85.8 | IQ3_XXS probably better |
57 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 90.5 | lower quality |
58 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-IQ3_XS.gguf) | i1-IQ3_XS | 96.1 | |
59 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-Q3_K_S.gguf) | i1-Q3_K_S | 101.5 | IQ3_XS probably better |
60 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-IQ3_S.gguf) | i1-IQ3_S | 101.6 | beats Q3_K* |
61 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-IQ3_M.gguf) | i1-IQ3_M | 103.2 | |
62 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-Q3_K_M.gguf) | i1-Q3_K_M | 112.5 | IQ3_S probably better |
63 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-Q3_K_L.gguf) | i1-Q3_K_L | 121.9 | IQ3_M probably better |
64 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-IQ4_XS.gguf) | i1-IQ4_XS | 125.4 | |
65 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-Q4_0.gguf) | i1-Q4_0 | 133.2 | fast, low quality |
66 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-Q4_K_S.gguf) | i1-Q4_K_S | 133.8 | optimal size/speed/quality |
67 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-Q4_K_M.gguf) | i1-Q4_K_M | 142.3 | fast, recommended |
68 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-Q4_1.gguf) | i1-Q4_1 | 147.3 | |
69 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-Q5_K_S.gguf) | i1-Q5_K_S | 162.0 | |
70 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-Q5_K_M.gguf) | i1-Q5_K_M | 166.9 | |
71 | [GGUF](https://huggingface.co/mradermacher/AReaL-SEA-235B-A22B-i1-GGUF/resolve/main/AReaL-SEA-235B-A22B.i1-Q6_K.gguf) | i1-Q6_K | 193.1 | practically like static Q6_K |
72
73 Here is a handy graph by ikawrakow comparing some lower-quality quant
74 types (lower is better):
75
76 ![image.png](https://www.nethype.de/huggingface_embed/quantpplgraph.png)
77
78 And here are Artefact2's thoughts on the matter:
79 https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
80
81 ## FAQ / Model Request
82
83 See https://huggingface.co/mradermacher/model_requests for some answers to
84 questions you might have and/or if you want some other model quantized.
85
86 ## Thanks
87
88 I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
89 me use its servers and providing upgrades to my workstation to enable
90 this work in my free time. Additional thanks to [@nicoboss](https://huggingface.co/nicoboss) for giving me access to his private supercomputer, enabling me to provide many more imatrix quants, at much higher quality, than I would otherwise be able to.
91
92 <!-- end -->
93