README.md
6.3 KB · 94 lines · markdown Raw
1 ---
2 base_model: Jarrodbarnes/KernelBench-RLVR-120b
3 datasets:
4 - ScalingIntelligence/KernelBench
5 language:
6 - en
7 library_name: transformers
8 license: apache-2.0
9 mradermacher:
10 readme_rev: 1
11 quantized_by: mradermacher
12 tags:
13 - gpu-kernel
14 - cuda
15 - code-generation
16 - reinforcement-learning
17 - grpo
18 - kernelbench
19 ---
20 ## About
21
22 <!-- ### quantize_version: 2 -->
23 <!-- ### output_tensor_quantised: 1 -->
24 <!-- ### convert_type: hf -->
25 <!-- ### vocab_type: -->
26 <!-- ### tags: nicoboss -->
27 <!-- ### 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 -->
28 <!-- ### quants_skip: -->
29 <!-- ### skip_mmproj: -->
30 weighted/imatrix quants of https://huggingface.co/Jarrodbarnes/KernelBench-RLVR-120b
31
32 <!-- provided-files -->
33
34 ***For a convenient overview and download list, visit our [model page for this model](https://hf.tst.eu/model#KernelBench-RLVR-120b-i1-GGUF).***
35
36 static quants are available at https://huggingface.co/mradermacher/KernelBench-RLVR-120b-GGUF
37 ## Usage
38
39 If you are unsure how to use GGUF files, refer to one of [TheBloke's
40 READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
41 more details, including on how to concatenate multi-part files.
42
43 ## Provided Quants
44
45 (sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
46
47 | Link | Type | Size/GB | Notes |
48 |:-----|:-----|--------:|:------|
49 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.imatrix.gguf) | imatrix | 0.3 | imatrix file (for creating your own quants) |
50 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-IQ1_M.gguf) | i1-IQ1_M | 66.2 | mostly desperate |
51 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-IQ1_S.gguf) | i1-IQ1_S | 66.2 | for the desperate |
52 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-IQ2_XXS.gguf) | i1-IQ2_XXS | 66.2 | |
53 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-IQ2_XS.gguf) | i1-IQ2_XS | 66.2 | |
54 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-Q3_K_S.gguf) | i1-Q3_K_S | 66.2 | IQ3_XS probably better |
55 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-IQ2_M.gguf) | i1-IQ2_M | 66.3 | |
56 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-IQ2_S.gguf) | i1-IQ2_S | 66.3 | |
57 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-IQ3_S.gguf) | i1-IQ3_S | 66.3 | beats Q3_K* |
58 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-IQ3_XS.gguf) | i1-IQ3_XS | 66.3 | |
59 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 66.3 | lower quality |
60 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-Q2_K.gguf) | i1-Q2_K | 66.3 | IQ3_XXS probably better |
61 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-IQ4_XS.gguf) | i1-IQ4_XS | 66.3 | |
62 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-Q4_0.gguf) | i1-Q4_0 | 66.6 | fast, low quality |
63 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-Q2_K_S.gguf) | i1-Q2_K_S | 66.7 | very low quality |
64 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-IQ3_M.gguf) | i1-IQ3_M | 66.8 | |
65 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-Q3_K_M.gguf) | i1-Q3_K_M | 71.2 | IQ3_S probably better |
66 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-Q3_K_L.gguf) | i1-Q3_K_L | 73.5 | IQ3_M probably better |
67 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-Q4_1.gguf) | i1-Q4_1 | 73.6 | |
68 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-Q4_K_S.gguf) | i1-Q4_K_S | 81.0 | optimal size/speed/quality |
69 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-Q4_K_M.gguf) | i1-Q4_K_M | 88.0 | fast, recommended |
70 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-Q5_K_S.gguf) | i1-Q5_K_S | 88.1 | |
71 | [GGUF](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-Q5_K_M.gguf) | i1-Q5_K_M | 94.0 | |
72 | [PART 1](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-Q6_K.gguf.part1of3) [PART 2](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-Q6_K.gguf.part2of3) [PART 3](https://huggingface.co/mradermacher/KernelBench-RLVR-120b-i1-GGUF/resolve/main/KernelBench-RLVR-120b.i1-Q6_K.gguf.part3of3) | i1-Q6_K | 124.3 | practically like static Q6_K |
73
74 Here is a handy graph by ikawrakow comparing some lower-quality quant
75 types (lower is better):
76
77 ![image.png](https://www.nethype.de/huggingface_embed/quantpplgraph.png)
78
79 And here are Artefact2's thoughts on the matter:
80 https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
81
82 ## FAQ / Model Request
83
84 See https://huggingface.co/mradermacher/model_requests for some answers to
85 questions you might have and/or if you want some other model quantized.
86
87 ## Thanks
88
89 I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
90 me use its servers and providing upgrades to my workstation to enable
91 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.
92
93 <!-- end -->
94