Neuratal commited on
Commit
c7930e0
·
verified ·
1 Parent(s): 569d682

Add model card (README pointing to source and purpose)

Browse files
Files changed (1) hide show
  1. README.md +65 -0
README.md ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - tr
4
+ - en
5
+ tags:
6
+ - transformer
7
+ - pytorch
8
+ - llm
9
+ - decoder-only
10
+ - neural-network
11
+ - gqa
12
+ - rope
13
+ - swiglu
14
+ - apache-2.0
15
+ - neuratal
16
+ pretty_name: Neuratal NTransformer v1.4.0
17
+ license: apache-2.0
18
+ ---
19
+
20
+ # NTransformer — Neuratal's open-source PyTorch Transformer
21
+
22
+ **NTransformer is not a pretrained model; this page documents the project, not model weights.**
23
+
24
+ NTransformer is an independently developed, PyTorch-based, decoder-only
25
+ Transformer research/engineering framework by Neuratal (https://neuratal.com).
26
+
27
+ - **Not pretrained.** No weights are distributed with the repository; outputs from an untrained preset are gibberish.
28
+ - **NTokenizer-native.** Vocab 262,144; token IDs consumed verbatim; SHA-256 fingerprint-verified checkpoints against the NTokenizer v0.3 artifact.
29
+ - **Apache-2.0.** NTokenizer is separate (CC-BY-SA-4.0); NTransformer is Apache-2.0.
30
+ - **Full NTokenizer integration:** the entire pipeline — token IDs consumed verbatim, SHA-256 fingerprint bound to checkpoints, no BOS/EOS injection, correct special-token map. **Designed to operate as the language model for NTokenizer v0.3-polished.**
31
+ - **Source + docs:** https://github.com/Neuratal/NTransformer
32
+ - **README:** https://github.com/Neuratal/NTransformer/blob/master/README.md
33
+ - **Latest release:** https://github.com/Neuratal/NTransformer/releases/tag/v1.4.0
34
+
35
+ ## What it is
36
+
37
+ A research/engineering framework covering architecture, inference, training,
38
+ and checkpointing — everything you need to train, evaluate and iterate on a
39
+ decoder-only Transformer without vendoring a tokenizer, model hub, or hosted
40
+ inference service. NTransformer is **not** a pretrained model; no trained
41
+ weights are distributed. No quality claims are made. Use as a research tool,
42
+ not as a production ML stack.
43
+
44
+ - **Architecture block:** decoder-only pre-norm; GQA, RoPE, SwiGLU, RMSNorm,
45
+ factorized tied embeddings — all LLaMA-style.
46
+ - **Long-context** paths for 16K/32K sequences with gradient checkpointing.
47
+ - **Tested:** 460 tests, real release, honest release notes.
48
+
49
+ ## Why this is here
50
+
51
+ NTransformer needs a presence on Hugging Face so that the project card, tags and references live where serious ML practitioners browse. This page intentionally does **not** host a model; it points to the open-source repository.
52
+
53
+ ## Links
54
+
55
+ - **GitHub repository:** https://github.com/Neuratal/NTransformer
56
+ - **NTokenizer (CC-BY-SA-4.0):** https://huggingface.co/Neuratal/ntokenizer
57
+ - **About Neuratal:** https://neuratal.com/about
58
+ - **License:** Apache-2.0
59
+
60
+ ## Limitations (honest)
61
+
62
+ - No trained weights.
63
+ - Not production support; research-oriented.
64
+ - Multi-GPU/NCCL not verified on the development hardware.
65
+ - Benchmarks are development-machine numbers; do not treat as broad-hardware measurements.