joellidin commited on
Commit
b8f8fd1
·
verified ·
1 Parent(s): ca09d35

Update model card

Browse files
Files changed (1) hide show
  1. README.md +44 -1
README.md CHANGED
@@ -2,4 +2,47 @@
2
  license: apache-2.0
3
  datasets:
4
  - mlfoundations/dclm-baseline-1.0-parquet
5
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2
  license: apache-2.0
3
  datasets:
4
  - mlfoundations/dclm-baseline-1.0-parquet
5
+ ---
6
+
7
+ # Covenant72B
8
+
9
+ **Covenant72B** is the largest permissionless collaboratively trained language
10
+ model trained entirely from scratch at the 72 billion parameter scale.
11
+
12
+ It is being trained with 20+ globally distributed participants coordinated via
13
+ decentralized infrastructure on the Bittensor blockchain.
14
+
15
+ **Checkpoint-One** marks the first release, corresponding to **200 billion
16
+ tokens processed**. Model files are available in the [Checkpoint-One
17
+ branch](https://huggingface.co/tplr/Covenant72B/tree/Checkpoint-One). Future
18
+ checkpoints will be updated here.
19
+
20
+ ---
21
+
22
+ ## Training Details
23
+
24
+ | Property | Value |
25
+ |-----------|--------|
26
+ | **Model size** | 72B |
27
+ | **Architecture** | LLaMA-style |
28
+ | **Target token budget** | 1.2T (210B for current checkpoint) |
29
+ | **Compute participants** | 20+ |
30
+ | **Minimal compute per participant** | 8×B200 or equivalent |
31
+ | **Dataset** | DCLM-baseline |
32
+ | **Optimizer** | SparseLoCo (communication-efficient optimizer) |
33
+
34
+ ---
35
+
36
+ ## Performance on Benchmarks
37
+ _All results are 0-shot acc-norm (%)_
38
+
39
+ | Model | Compute Environment / Permissions | Size | Tokens | ARC-C | ARC-E | PIQA | OpenBookQA | HellaSwag | Winogrande | MMLU |
40
+ |:------|:----------------------------------|------:|--------:|------:|------:|------:|------------:|-----------:|-------------:|------:|
41
+ | **Intellect-1** | Over the internet / White List | 10B | 1T | 44.8 | 71.6 | 77.7 | 43.6 | 70.5 | 63.1 | 32.7 |
42
+ | **Psyche Consilience-7Y9** | Over the internet / White List | 40B | 1.2T | 31.1 | 55.8 | 76.1 | 34.8 | 63.7 | 57.0 | 24.2 |
43
+ | **Covenant72B – Checkpoint One** | Over the internet / Permissionless | 70B | 210B | 46.2 | 72.6 | 79.2 | 43.0 | 73.5 | 70.3 | 38.0 |
44
+ | **K2 Checkpoint 54** | Centralized Cluster | 65B | 210B | 41.8 | 69.5 | 80.1 | 42.4 | 74.9 | 68.9 | 33.7 |
45
+
46
+ ---
47
+
48
+ For more details, refer to [Checkpoint One on Templar Research](https://templarresearch.substack.com/p/checkpoint-one).