Simplify model card artifact details
Browse files- README.md +5 -26
- SHA256SUMS +1 -1
README.md
CHANGED
|
@@ -40,27 +40,6 @@ This repository provides the same final task-specific DiBO model in two formats.
|
|
| 40 |
The safetensors model was not trained separately. The LLaDA base weights are
|
| 41 |
not duplicated in this repository.
|
| 42 |
|
| 43 |
-
## Artifact details
|
| 44 |
-
|
| 45 |
-
- Base model: [`GSAI-ML/LLaDA-8B-Instruct`](https://huggingface.co/GSAI-ML/LLaDA-8B-Instruct), revision `08b83a6feb34df1a6011b80c3c00c7563e963b07`
|
| 46 |
-
- Original checkpoint: `dibo_tfbind8_final.pt` (16,030,379,937 bytes; SHA-256 `0d8e17f7aa60baf4173d71a4ec8475d182a992812b59d29a728eec719d1b67bf`)
|
| 47 |
-
- Original format: PyTorch checkpoint dictionary, state dict at `model`
|
| 48 |
-
- Transformers model class: `DiBOLLaDAModelLM` loaded with `AutoModel`
|
| 49 |
-
- Tokenizer: `PreTrainedTokenizerFast`, vocabulary size `126,353`
|
| 50 |
-
- Input/output vocabulary rows: `126,353` / `126,464`
|
| 51 |
-
- Required delimiter tokens: `|design-start|`, `|design-end|`, `|label-start|`, `|label-end|`
|
| 52 |
-
|
| 53 |
-
The standard export contains the following safetensors shards:
|
| 54 |
-
- `model-00001-of-00004.safetensors` (4,994,672,360 bytes; SHA-256 `b1d0d9d09be47f1c47dff91926fd7f3eda272dc2ecbdd1b1a68f4bf42984473c`)
|
| 55 |
-
- `model-00002-of-00004.safetensors` (4,999,819,552 bytes; SHA-256 `dbe8101a92f660a2c2bd3d5d3dae6cc58f8258ccafc5d487d627f4aa7f48f7ad`)
|
| 56 |
-
- `model-00003-of-00004.safetensors` (4,999,802,728 bytes; SHA-256 `62c12ec9f508f42811eaa1880dfcb7c4fbca1a48d8379e796d2e8f34f978abfe`)
|
| 57 |
-
- `model-00004-of-00004.safetensors` (1,035,993,232 bytes; SHA-256 `0a6587d94453234ffa5d0e18cddfde93219c40754a2d5359e0ce1e095f714286`)
|
| 58 |
-
|
| 59 |
-
Its total weight size is `16,030,287,872` bytes. The
|
| 60 |
-
machine-readable [artifact manifest](artifact_manifest.json) and
|
| 61 |
-
[validation report](export_validation.json) record source hashes and measured
|
| 62 |
-
validation results. Validation passed for: candidate parity, config equality, forward parity, remote from pretrained, remote pt download, strict load, task smoke, tokenizer equality, weight equality.
|
| 63 |
-
|
| 64 |
## A. Load the standard Transformers export
|
| 65 |
|
| 66 |
```python
|
|
@@ -69,12 +48,12 @@ from transformers import AutoModel, AutoTokenizer
|
|
| 69 |
repo_id = "zpointsun/DiBO-TFBind8"
|
| 70 |
tokenizer = AutoTokenizer.from_pretrained(
|
| 71 |
repo_id,
|
| 72 |
-
revision="v1.1.
|
| 73 |
trust_remote_code=True,
|
| 74 |
)
|
| 75 |
model = AutoModel.from_pretrained(
|
| 76 |
repo_id,
|
| 77 |
-
revision="v1.1.
|
| 78 |
trust_remote_code=True,
|
| 79 |
use_safetensors=True,
|
| 80 |
torch_dtype="auto",
|
|
@@ -93,7 +72,7 @@ and strictly loads `checkpoint["model"]`.
|
|
| 93 |
|
| 94 |
```bash
|
| 95 |
hf download zpointsun/DiBO-TFBind8 dibo_tfbind8_final.pt \
|
| 96 |
-
--revision v1.1.
|
| 97 |
```
|
| 98 |
|
| 99 |
```python
|
|
@@ -106,7 +85,7 @@ assert LLADA_MODEL_REVISION == "08b83a6feb34df1a6011b80c3c00c7563e963b07"
|
|
| 106 |
checkpoint_path = hf_hub_download(
|
| 107 |
"zpointsun/DiBO-TFBind8",
|
| 108 |
filename="dibo_tfbind8_final.pt",
|
| 109 |
-
revision="v1.1.
|
| 110 |
)
|
| 111 |
model, tokenizer = load_model_and_tokenizer(DEFAULT_MODEL_ID, device="cuda")
|
| 112 |
checkpoint = torch.load(checkpoint_path, map_location="cuda")
|
|
@@ -121,7 +100,7 @@ From a checkout of the released DiBO code and its oracle environment:
|
|
| 121 |
```bash
|
| 122 |
# Standard Transformers export
|
| 123 |
python eval.py --tasks TFBind8-Exact-v0 \
|
| 124 |
-
--model_name_or_path zpointsun/DiBO-TFBind8 --model_revision v1.1.
|
| 125 |
--seeds <SEEDS> --max_attempts 1000
|
| 126 |
|
| 127 |
# Canonical local .pt checkpoint
|
|
|
|
| 40 |
The safetensors model was not trained separately. The LLaDA base weights are
|
| 41 |
not duplicated in this repository.
|
| 42 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
## A. Load the standard Transformers export
|
| 44 |
|
| 45 |
```python
|
|
|
|
| 48 |
repo_id = "zpointsun/DiBO-TFBind8"
|
| 49 |
tokenizer = AutoTokenizer.from_pretrained(
|
| 50 |
repo_id,
|
| 51 |
+
revision="v1.1.2",
|
| 52 |
trust_remote_code=True,
|
| 53 |
)
|
| 54 |
model = AutoModel.from_pretrained(
|
| 55 |
repo_id,
|
| 56 |
+
revision="v1.1.2",
|
| 57 |
trust_remote_code=True,
|
| 58 |
use_safetensors=True,
|
| 59 |
torch_dtype="auto",
|
|
|
|
| 72 |
|
| 73 |
```bash
|
| 74 |
hf download zpointsun/DiBO-TFBind8 dibo_tfbind8_final.pt \
|
| 75 |
+
--revision v1.1.2 --local-dir checkpoints/dibo-tfbind8
|
| 76 |
```
|
| 77 |
|
| 78 |
```python
|
|
|
|
| 85 |
checkpoint_path = hf_hub_download(
|
| 86 |
"zpointsun/DiBO-TFBind8",
|
| 87 |
filename="dibo_tfbind8_final.pt",
|
| 88 |
+
revision="v1.1.2",
|
| 89 |
)
|
| 90 |
model, tokenizer = load_model_and_tokenizer(DEFAULT_MODEL_ID, device="cuda")
|
| 91 |
checkpoint = torch.load(checkpoint_path, map_location="cuda")
|
|
|
|
| 100 |
```bash
|
| 101 |
# Standard Transformers export
|
| 102 |
python eval.py --tasks TFBind8-Exact-v0 \
|
| 103 |
+
--model_name_or_path zpointsun/DiBO-TFBind8 --model_revision v1.1.2 \
|
| 104 |
--seeds <SEEDS> --max_attempts 1000
|
| 105 |
|
| 106 |
# Canonical local .pt checkpoint
|
SHA256SUMS
CHANGED
|
@@ -1,4 +1,4 @@
|
|
| 1 |
-
|
| 2 |
ffaa3217d5f7d83a027fca9df0070c0082726c6770f44bb94f1caf931036539b UPSTREAM_NOTICE.md
|
| 3 |
ea63d0120e6dac7cfdb9ded5e2f4f9716092ee6f8e1b416b65aa60a8e66164f8 __init__.py
|
| 4 |
b60301d77b08cf9e754154757182ed3e4ae255bc189192d3aaa0969ec1969888 artifact_manifest.json
|
|
|
|
| 1 |
+
1cbf76a1ee8e4c060e68c5726aa0ac96407a5dc940bd1aee03d07f2b97dad19d README.md
|
| 2 |
ffaa3217d5f7d83a027fca9df0070c0082726c6770f44bb94f1caf931036539b UPSTREAM_NOTICE.md
|
| 3 |
ea63d0120e6dac7cfdb9ded5e2f4f9716092ee6f8e1b416b65aa60a8e66164f8 __init__.py
|
| 4 |
b60301d77b08cf9e754154757182ed3e4ae255bc189192d3aaa0969ec1969888 artifact_manifest.json
|