IMvision12 commited on
Commit
6b82020
·
verified ·
1 Parent(s): 4c07a2f

Fix pipeline_tag to a valid HF Hub tag

Browse files
Files changed (1) hide show
  1. README.md +94 -95
README.md CHANGED
@@ -1,95 +1,94 @@
1
- ---
2
- pipeline_tag: text2text-generation
3
- license: apache-2.0
4
- base_model: facebook/bart-large
5
- library_name: zeromodels
6
- tags:
7
- - keras
8
- - zeromodels
9
- - bart
10
- - text2text-generation
11
- - seq2seq
12
- - arxiv:1910.13461
13
- - pytorch
14
- - jax
15
- - tf
16
- ---
17
-
18
- # Run BART with Keras 3: JAX, PyTorch, or TensorFlow
19
-
20
- [![GitHub](https://img.shields.io/badge/GitHub-ZeroModels-black?logo=github)](https://github.com/ZeroAIx/ZeroModels) [![Docs](https://img.shields.io/badge/Docs-BART-blue)](https://zeroaix.github.io/ZeroModels/bart/)
21
-
22
- # zeromodels/bart_large
23
-
24
- Paper: [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension (arXiv:1910.13461)](https://arxiv.org/abs/1910.13461) · [HF Papers](https://huggingface.co/papers/1910.13461)
25
-
26
- BART is a denoising seq2seq transformer: a bidirectional encoder (like BERT) and an autoregressive decoder (like GPT) trained to reconstruct corrupted text. It excels at summarization, translation, and other text-to-text tasks. Byte-level BPE tokenizer (shared with RoBERTa); the decoder starts from `</s>`.
27
-
28
- For more details on the model, please go to the upstream [model card](https://huggingface.co/facebook/bart-large).
29
-
30
- Pure-**Keras 3** conversion of [`facebook/bart-large`](https://huggingface.co/facebook/bart-large) for [zeromodels](https://github.com/ZeroAIx/ZeroModels). One implementation runs unmodified on **TensorFlow / Torch / JAX**.
31
-
32
- This is a **conditional generation (base seq2seq)** checkpoint (`BartConditionalGenerate`). Other task heads load the shared backbone from this repo (start randomly initialized, ready for fine-tuning); fine-tuned task checkpoints load via the `hf:` prefix.
33
-
34
- > Base checkpoint (not task fine-tuned): use it as a backbone (`BartModel`) for features, or fine-tune a task head.
35
-
36
- ## ✨ Quick start
37
-
38
- ```python
39
- import os
40
- os.environ["KERAS_BACKEND"] = "torch" # or "jax" / "tensorflow"
41
-
42
- from zeromodels.models.bart import BartConditionalGenerate, BartTokenizer
43
-
44
- model = BartConditionalGenerate.from_weights("zeromodels/bart_large")
45
- tokenizer = BartTokenizer.from_weights("zeromodels/bart_large")
46
-
47
- inputs = tokenizer('The quick brown fox jumps over the lazy dog.')
48
- ids = model.generate(
49
- inputs,
50
- [[model.decoder_start_token_id]],
51
- max_new_tokens=64,
52
- eos_token_id=tokenizer.eos_token_id,
53
- )
54
- print(tokenizer.decode(ids[0], skip_special_tokens=True))
55
- ```
56
-
57
- Load any BART variant the same way with `from_weights("zeromodels/<variant>")`:
58
-
59
- | Variant | Hub | Task |
60
- |---|---|---|
61
- | `bart_base` | [`zeromodels/bart_base`](https://huggingface.co/zeromodels/bart_base) | conditional generation (base seq2seq) |
62
- | `bart_large` | [`zeromodels/bart_large`](https://huggingface.co/zeromodels/bart_large) | conditional generation (base seq2seq) |
63
- | `bart_large_cnn` | [`zeromodels/bart_large_cnn`](https://huggingface.co/zeromodels/bart_large_cnn) | summarization (CNN / DailyMail) |
64
- | `bart_large_xsum` | [`zeromodels/bart_large_xsum`](https://huggingface.co/zeromodels/bart_large_xsum) | extreme summarization (XSum, one-sentence) |
65
-
66
- ## Available classes
67
-
68
- Load any of these from this repo with `from_weights("zeromodels/bart_large")` (or on the fly via the `hf:` prefix). The pretrained backbone is shared; task heads not stored in this checkpoint start randomly initialized, ready for fine-tuning (or load a `hf:` fine-tune).
69
-
70
- | Class | Task |
71
- |---|---|
72
- | `BartModel` | Encoder-decoder backbone |
73
- | `BartConditionalGenerate` | Conditional generation (summarization / seq2seq) |
74
- | `BartSequenceClassify` | Sequence classification (e.g. NLI / zero-shot) |
75
- | `BartQnA` | Extractive question answering |
76
-
77
- ```python
78
- from zeromodels.models.bart import BartSequenceClassify
79
- # zero-shot / NLI fine-tune loads on the fly via the hf: prefix
80
- model = BartSequenceClassify.from_weights("hf:facebook/bart-large-mnli")
81
- ```
82
-
83
- ## Tips
84
-
85
- - Set `KERAS_BACKEND` **before** importing Keras / zeromodels.
86
- - Prefer `BartTokenizer.from_weights(...)` so the byte-level BPE matches.
87
- - BART's decoder starts from `</s>` (`decoder_start_token_id = 2`); pass `eos_token_id=tokenizer.eos_token_id` to stop generation.
88
- - See the [BART docs](https://zeroaix.github.io/ZeroModels/bart/) and [Loading Weights](https://zeroaix.github.io/ZeroModels/loading_weights/).
89
- - Community / upstream safetensors still work via the `hf:` prefix, e.g. `BartConditionalGenerate.from_weights("hf:facebook/bart-large")`.
90
-
91
- ## Special Thanks
92
-
93
- A huge thank you to the Meta AI (FAIR) authors for creating and releasing BART.
94
-
95
- License: apache-2.0.
 
1
+ ---
2
+ pipeline_tag: feature-extraction
3
+ license: apache-2.0
4
+ base_model: facebook/bart-large
5
+ library_name: zeromodels
6
+ tags:
7
+ - keras
8
+ - zeromodels
9
+ - bart
10
+ - seq2seq
11
+ - arxiv:1910.13461
12
+ - pytorch
13
+ - jax
14
+ - tf
15
+ ---
16
+
17
+ # Run BART with Keras 3: JAX, PyTorch, or TensorFlow
18
+
19
+ [![GitHub](https://img.shields.io/badge/GitHub-ZeroModels-black?logo=github)](https://github.com/ZeroAIx/ZeroModels) [![Docs](https://img.shields.io/badge/Docs-BART-blue)](https://zeroaix.github.io/ZeroModels/bart/)
20
+
21
+ # zeromodels/bart_large
22
+
23
+ Paper: [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension (arXiv:1910.13461)](https://arxiv.org/abs/1910.13461) · [HF Papers](https://huggingface.co/papers/1910.13461)
24
+
25
+ BART is a denoising seq2seq transformer: a bidirectional encoder (like BERT) and an autoregressive decoder (like GPT) trained to reconstruct corrupted text. It excels at summarization, translation, and other text-to-text tasks. Byte-level BPE tokenizer (shared with RoBERTa); the decoder starts from `</s>`.
26
+
27
+ For more details on the model, please go to the upstream [model card](https://huggingface.co/facebook/bart-large).
28
+
29
+ Pure-**Keras 3** conversion of [`facebook/bart-large`](https://huggingface.co/facebook/bart-large) for [zeromodels](https://github.com/ZeroAIx/ZeroModels). One implementation runs unmodified on **TensorFlow / Torch / JAX**.
30
+
31
+ This is a **conditional generation (base seq2seq)** checkpoint (`BartConditionalGenerate`). Other task heads load the shared backbone from this repo (start randomly initialized, ready for fine-tuning); fine-tuned task checkpoints load via the `hf:` prefix.
32
+
33
+ > Base checkpoint (not task fine-tuned): use it as a backbone (`BartModel`) for features, or fine-tune a task head.
34
+
35
+ ## ✨ Quick start
36
+
37
+ ```python
38
+ import os
39
+ os.environ["KERAS_BACKEND"] = "torch" # or "jax" / "tensorflow"
40
+
41
+ from zeromodels.models.bart import BartConditionalGenerate, BartTokenizer
42
+
43
+ model = BartConditionalGenerate.from_weights("zeromodels/bart_large")
44
+ tokenizer = BartTokenizer.from_weights("zeromodels/bart_large")
45
+
46
+ inputs = tokenizer('The quick brown fox jumps over the lazy dog.')
47
+ ids = model.generate(
48
+ inputs,
49
+ [[model.decoder_start_token_id]],
50
+ max_new_tokens=64,
51
+ eos_token_id=tokenizer.eos_token_id,
52
+ )
53
+ print(tokenizer.decode(ids[0], skip_special_tokens=True))
54
+ ```
55
+
56
+ Load any BART variant the same way with `from_weights("zeromodels/<variant>")`:
57
+
58
+ | Variant | Hub | Task |
59
+ |---|---|---|
60
+ | `bart_base` | [`zeromodels/bart_base`](https://huggingface.co/zeromodels/bart_base) | conditional generation (base seq2seq) |
61
+ | `bart_large` | [`zeromodels/bart_large`](https://huggingface.co/zeromodels/bart_large) | conditional generation (base seq2seq) |
62
+ | `bart_large_cnn` | [`zeromodels/bart_large_cnn`](https://huggingface.co/zeromodels/bart_large_cnn) | summarization (CNN / DailyMail) |
63
+ | `bart_large_xsum` | [`zeromodels/bart_large_xsum`](https://huggingface.co/zeromodels/bart_large_xsum) | extreme summarization (XSum, one-sentence) |
64
+
65
+ ## Available classes
66
+
67
+ Load any of these from this repo with `from_weights("zeromodels/bart_large")` (or on the fly via the `hf:` prefix). The pretrained backbone is shared; task heads not stored in this checkpoint start randomly initialized, ready for fine-tuning (or load a `hf:` fine-tune).
68
+
69
+ | Class | Task |
70
+ |---|---|
71
+ | `BartModel` | Encoder-decoder backbone |
72
+ | `BartConditionalGenerate` | Conditional generation (summarization / seq2seq) |
73
+ | `BartSequenceClassify` | Sequence classification (e.g. NLI / zero-shot) |
74
+ | `BartQnA` | Extractive question answering |
75
+
76
+ ```python
77
+ from zeromodels.models.bart import BartSequenceClassify
78
+ # zero-shot / NLI fine-tune loads on the fly via the hf: prefix
79
+ model = BartSequenceClassify.from_weights("hf:facebook/bart-large-mnli")
80
+ ```
81
+
82
+ ## Tips
83
+
84
+ - Set `KERAS_BACKEND` **before** importing Keras / zeromodels.
85
+ - Prefer `BartTokenizer.from_weights(...)` so the byte-level BPE matches.
86
+ - BART's decoder starts from `</s>` (`decoder_start_token_id = 2`); pass `eos_token_id=tokenizer.eos_token_id` to stop generation.
87
+ - See the [BART docs](https://zeroaix.github.io/ZeroModels/bart/) and [Loading Weights](https://zeroaix.github.io/ZeroModels/loading_weights/).
88
+ - Community / upstream safetensors still work via the `hf:` prefix, e.g. `BartConditionalGenerate.from_weights("hf:facebook/bart-large")`.
89
+
90
+ ## Special Thanks
91
+
92
+ A huge thank you to the Meta AI (FAIR) authors for creating and releasing BART.
93
+
94
+ License: apache-2.0.