Add model card listing per-step branches
Browse files
README.md
ADDED
|
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- checkpoint
|
| 4 |
+
- pretraining
|
| 5 |
+
- climbmix
|
| 6 |
+
---
|
| 7 |
+
|
| 8 |
+
# CloverLM-8b — training checkpoints
|
| 9 |
+
|
| 10 |
+
Raw PyTorch training checkpoints for CloverLM-8b (climbmix run). **Each training
|
| 11 |
+
step lives on its own git branch** so you can download exactly the checkpoint you
|
| 12 |
+
need without pulling the whole history.
|
| 13 |
+
|
| 14 |
+
| Training step | Branch | File |
|
| 15 |
+
|---:|---|---|
|
| 16 |
+
| 20000 | `step-20000` | `climbmix-8b-20000.pt` |
|
| 17 |
+
| 40000 | `step-40000` | `climbmix-8b-40000.pt` |
|
| 18 |
+
| 60000 | `step-60000` | `climbmix-8b-60000.pt` |
|
| 19 |
+
| 80000 | `step-80000` | `climbmix-8b-80000.pt` |
|
| 20 |
+
| 100000 | `step-100000` | `climbmix-8b-100000.pt` |
|
| 21 |
+
| 120000 | `step-120000` | `climbmix-8b-120000.pt` |
|
| 22 |
+
| 140000 | `step-140000` | `climbmix-8b-140000.pt` |
|
| 23 |
+
| 160000 | `step-160000` | `climbmix-8b-160000.pt` |
|
| 24 |
+
| 165000 | `step-165000` | `climbmix-8b-165000.pt` |
|
| 25 |
+
| 180000 | `step-180000` | `climbmix-8b-180000.pt` |
|
| 26 |
+
| 200000 | `step-200000` | `climbmix-8b-200000.pt` |
|
| 27 |
+
| 220000 | `step-220000` | `climbmix-8b-220000.pt` |
|
| 28 |
+
| 240000 | `step-240000` | `climbmix-8b-240000.pt` |
|
| 29 |
+
| 250000 | `step-250000` | `climbmix-8b-250000.pt` |
|
| 30 |
+
| 260000 | `step-260000` | `climbmix-8b-260000.pt` |
|
| 31 |
+
| 280000 | `step-280000` | `climbmix-8b-280000.pt` |
|
| 32 |
+
| 300000 | `step-300000` | `climbmix-8b-300000.pt` |
|
| 33 |
+
| 320000 | `step-320000` | `climbmix-8b-320000.pt` |
|
| 34 |
+
| 340000 | `step-340000` | `climbmix-8b-340000.pt` |
|
| 35 |
+
| 360000 | `step-360000` | `climbmix-8b-360000.pt` |
|
| 36 |
+
|
| 37 |
+
## Download one checkpoint
|
| 38 |
+
|
| 39 |
+
```python
|
| 40 |
+
from huggingface_hub import hf_hub_download
|
| 41 |
+
import torch
|
| 42 |
+
|
| 43 |
+
path = hf_hub_download(
|
| 44 |
+
repo_id="daslab-testing/CloverLM-8b",
|
| 45 |
+
filename="climbmix-8b-20000.pt",
|
| 46 |
+
revision="step-20000", # pick the branch = training step you want
|
| 47 |
+
)
|
| 48 |
+
state = torch.load(path, map_location="cpu")
|
| 49 |
+
```
|
| 50 |
+
|
| 51 |
+
Or with the CLI:
|
| 52 |
+
|
| 53 |
+
```bash
|
| 54 |
+
hf download daslab-testing/CloverLM-8b climbmix-8b-20000.pt --revision step-20000 --local-dir .
|
| 55 |
+
```
|