Humuhumu33 commited on
Commit
1f1aed9
verified
1 Parent(s): 106a245

Add branded Hologram model card

Browse files
Files changed (1) hide show
  1. README.md +84 -0
README.md ADDED
@@ -0,0 +1,84 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ base_model: tiiuae/Falcon-E-3B-Instruct-GGUF
4
+ pipeline_tag: text-generation
5
+ inference: false
6
+ tags:
7
+ - hologram
8
+ - q
9
+ - kappa-object
10
+ - holo
11
+ - ternary
12
+ - bitnet
13
+ - falcon
14
+ ---
15
+
16
+ <div align="center">
17
+
18
+ # Hologram 路 Falcon-E-3B
19
+
20
+ **Ternary reasoning brain**
21
+
22
+ `t2 路 1.58-bit ternary` 路 `0.63 GB` 路 streamed to Q as a **key-addressable `.holo` object**
23
+
24
+ [Hologram](https://gethologram.ai) 路 [Live Space](https://huggingface.co/spaces/HOLOGRAMTECH/hologram) 路 [Organization](https://huggingface.co/HOLOGRAMTECH) 路 [Code](https://github.com/Hologram-Technologies)
25
+
26
+ </div>
27
+
28
+ ---
29
+
30
+ ## What this is
31
+
32
+ A 3B ternary instruct model, re-encoded to Hologram's 1.58-bit key format. Small enough to stream cold, strong enough to hold a conversation.
33
+
34
+ This repository is **not** a GGUF or Transformers checkpoint. It is a **Hologram key object**: the weights of `tiiuae/Falcon-E-3B-Instruct-GGUF` re-encoded into Hologram's content-addressed `.holo` format so they stream, one verified block at a time, into **Q**, the on-device brain of the Hologram web OS. It runs in the browser on WebGPU, serverless, with nothing to install.
35
+
36
+ ## How it streams
37
+
38
+ The object is laid out for cold streaming from an untrusted CDN:
39
+
40
+ | File | Role |
41
+ |---|---|
42
+ | `manifest.json` | the root. Names every tensor and the key (content hash) of its block. |
43
+ | `b/sha256_*.gz` | the tensor blocks. Each filename **is** the SHA-256 of its bytes. |
44
+ | `tokenizer.gguf` | bundled header (where present), so loading is fully serverless. |
45
+
46
+ Q fetches the manifest, then pulls each block by its key and re-derives `sha256(block)` on arrival. If a byte is wrong, the block is rejected. Nothing is trusted; everything is proven.
47
+
48
+ ## Verify (Law L5)
49
+
50
+ The object's identity is the SHA-256 of its manifest, pinned in Q's catalog before a single byte of weight is trusted:
51
+
52
+ ```
53
+ did:holo:sha256:6b753fe8186f2b4194424115c36014698580a2aab8427e9b40365893ac6b77ca
54
+ ```
55
+
56
+ ```bash
57
+ # the manifest hash equals the pinned identity above
58
+ curl -sL https://huggingface.co/HOLOGRAMTECH/q-falcon-e-3b/resolve/main/manifest.json | sha256sum
59
+ ```
60
+
61
+ ## Specifications
62
+
63
+ | | |
64
+ |---|---|
65
+ | Architecture | Falcon-E |
66
+ | Precision | t2 路 1.58-bit ternary |
67
+ | Object size | 0.63 GB |
68
+ | Hidden size | 2048 |
69
+ | Layers | 32 |
70
+ | Heads (Q / KV) | 16 / 2 (GQA) |
71
+ | FFN | 13312 |
72
+ | Vocab | 32768 |
73
+ | Context | 3000 |
74
+ | Format | `holo-2bit/1` |
75
+
76
+ ## Provenance and license
77
+
78
+ Derived from [`tiiuae/Falcon-E-3B-Instruct-GGUF`](https://huggingface.co/tiiuae/Falcon-E-3B-Instruct-GGUF). Inherits the Falcon-E license from tiiuae/Falcon-E-3B-Instruct. The re-encoding is lossless-by-construction at the key level: every block is content-addressed, so the object either re-derives to its pinned identity or it is refused.
79
+
80
+ ## Run it
81
+
82
+ These weights load through Q, not a standard runtime. Open the [Live Space](https://huggingface.co/spaces/HOLOGRAMTECH/hologram) or visit [gethologram.ai](https://gethologram.ai) to run Hologram, then pick **Falcon-E-3B** from Q's model list.
83
+
84
+ <div align="center"><sub>Composed on the golden ratio. One key, everything.</sub></div>