ashxhart commited on
Commit
40a1398
·
verified ·
1 Parent(s): 0728ee6

Rebrand organization card to TensorFold

Browse files
Files changed (4) hide show
  1. .gitattributes +2 -0
  2. README.md +23 -38
  3. tensorfold-hero.png +3 -0
  4. tensorfold-logo.png +3 -0
.gitattributes CHANGED
@@ -34,3 +34,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
  assets/vontra-logo-512.png filter=lfs diff=lfs merge=lfs -text
 
 
 
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
  assets/vontra-logo-512.png filter=lfs diff=lfs merge=lfs -text
37
+ tensorfold-logo.png filter=lfs diff=lfs merge=lfs -text
38
+ tensorfold-hero.png filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -1,62 +1,47 @@
1
  ---
2
- title: Vontra
3
  sdk: static
4
  pinned: false
5
  ---
6
 
7
-
8
  <div align="center">
 
9
 
 
10
 
11
- # Vontra
12
-
13
-
14
- ### Exploring local AI and sharing useful MLX work with the community.
15
-
16
 
 
17
  </div>
18
 
 
19
 
20
- Vontra is my personal local-AI project. I’m one person experimenting with open models, learning what today’s hardware can do, and publishing anything that might save somebody else time.
21
-
22
-
23
- My main focus is **MLX and Apple silicon**. I run models on a Mac Studio and MacBook Pro, work through conversion and compatibility problems, and share usable checkpoints, code and notes when I get them running. I also use a ThinkStation PGX (DGX Spark) for local inference experiments outside the Apple ecosystem.
24
-
25
-
26
- ## What you’ll find here
27
-
28
-
29
- - MLX conversions and quantized checkpoints
30
- - Compatibility work for architectures that need more than a stock conversion
31
- - Practical notes from running and serving models locally
32
- - Memory and performance observations where they are useful
33
- - Clear links and credit to the original model creators
34
- - Honest warnings when something is experimental or incomplete
35
-
36
-
37
- ## Why I share it
38
-
39
-
40
- Local AI moves quickly, and a lot of the difficult work happens between a model being released and somebody being able to run it well on their own machine. Sharing that work makes the ecosystem more accessible and helps the next person avoid solving the same problems again.
41
-
42
-
43
- This is a personal, best-effort project rather than a company or research lab. I don’t train the base models, and every release remains the work of its original authors. Each repository should explain what was changed, how it was tested, what needs to be trusted, and which upstream licence applies.
44
 
 
 
 
 
 
45
 
46
- If something here helps, use it. If something breaks, a clear and reproducible report is always appreciated.
47
 
 
 
 
48
 
 
49
 
50
  ## Choose a model for your Mac
51
 
52
  | Unified memory | Collection | Selection basis |
53
  | --- | --- | --- |
54
- | 64GB | [Browse models](https://huggingface.co/collections/Vontra/mlx-models-for-64gb-macs) | Published M3 Studio peak below 48 GB |
55
- | 128GB | [Browse models](https://huggingface.co/collections/Vontra/mlx-models-for-128gb-macs) | Published M3 Studio peak below 96 GB |
56
- | 256GB | [Browse models](https://huggingface.co/collections/Vontra/mlx-models-for-256gb-macs) | Published M3 Studio peak below 192 GB |
57
 
58
- These are memory-based starting points with at least 25% nominal headroom, not measurements on every Mac or guarantees at full context. Smaller-tier models can also be used on larger-memory systems. Each collection note links back to the original benchmark conditions; start with short context and a single request.
59
 
60
- The model cards include setup instructions, published speed and memory evidence where available, and a short demo prompt to try. Missing oMLX version or recorded-demo evidence is labelled explicitly.
61
 
62
- [Follow Vontra for new Apple Silicon releases and fixes.](https://huggingface.co/Vontra)
 
1
  ---
2
+ title: TensorFold
3
  sdk: static
4
  pinned: false
5
  ---
6
 
 
7
  <div align="center">
8
+ <img src="./tensorfold-hero.png" alt="TensorFold" width="100%">
9
 
10
+ # TensorFold
11
 
12
+ ### Fast local LLM inference and tested model releases for Apple Silicon and NVIDIA CUDA.
 
 
 
 
13
 
14
+ [tensorfold.dev](https://tensorfold.dev) · [Runtime and setup](https://github.com/ashhart/TensorFold) · [Model collections](https://huggingface.co/TensorFold/collections)
15
  </div>
16
 
17
+ TensorFold is an open-source inference runtime and a practical model-release project. The runtime serves local models behind an OpenAI-compatible API, while this Hugging Face organization publishes checkpoints and supporting assets tested on real hardware.
18
 
19
+ ## What you will find here
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
20
 
21
+ - MLX quantized checkpoints for Apple Silicon
22
+ - MTP and DFlash assets where the upstream model provides a compatible drafter
23
+ - NVIDIA and DGX Spark recipes when a release has been tested there
24
+ - Measured speed, memory use, runtime versions, and known limits
25
+ - Clear credit, licences, and links to the original model authors
26
 
27
+ ## Run with TensorFold
28
 
29
+ ```bash
30
+ curl -fsSL https://tensorfold.dev/install.sh | sh
31
+ ```
32
 
33
+ TensorFold supports macOS and Linux. See the [setup guide](https://github.com/ashhart/TensorFold) for current model families, runtime flags, and benchmark conditions.
34
 
35
  ## Choose a model for your Mac
36
 
37
  | Unified memory | Collection | Selection basis |
38
  | --- | --- | --- |
39
+ | 64 GB | [Browse models](https://huggingface.co/collections/TensorFold/mlx-models-for-64gb-macs) | Published peak below 48 GB |
40
+ | 128 GB | [Browse models](https://huggingface.co/collections/TensorFold/mlx-models-for-128gb-macs) | Published peak below 96 GB |
41
+ | 256 GB | [Browse models](https://huggingface.co/collections/TensorFold/mlx-models-for-256gb-macs) | Published peak below 192 GB |
42
 
43
+ These collections are starting points with nominal context headroom, not guarantees at maximum context. Each model card records the tested runtime, prompt, output length, memory evidence, and any compatibility caveats.
44
 
45
+ TensorFold does not train the base models. Model design, training, evaluations, and upstream documentation remain the work of the original authors and contributors.
46
 
47
+ [Follow TensorFold for new model releases, runtime updates, and fixes.](https://huggingface.co/TensorFold)
tensorfold-hero.png ADDED

Git LFS Details

  • SHA256: 5ed0d5370027575cdcbfb927af2856d06de4278f62557b615fbaa28c8d24c161
  • Pointer size: 132 Bytes
  • Size of remote file: 1.89 MB
tensorfold-logo.png ADDED

Git LFS Details

  • SHA256: 24361d503f48e2e833743dd95752cea1ae0b03850ed84654f0d938b258308cc6
  • Pointer size: 132 Bytes
  • Size of remote file: 1.03 MB