TieuDaoChanNhan commited on
Commit
b563df3
·
verified ·
1 Parent(s): d6b92ba

Upload VLA tokenizer with model card

Browse files
Files changed (2) hide show
  1. README.md +7 -5
  2. tokenizer.json +2 -2
README.md CHANGED
@@ -14,9 +14,10 @@ license: apache-2.0
14
  # VLA Tokenizer — Qwen3
15
 
16
  Extended Qwen3 tokenizer for the **FineVideo-VLA** multimodal dataset.
17
- Adds all VLA tokens (video, 3D pose, SNAC audio) on top of the Qwen3 base tokenizer.
 
18
 
19
- **Vocab size: 257,897** (~151,669 Qwen3 base + 106,228 VLA tokens)
20
 
21
  > **Why Qwen3?** The Qwen3 family has strong multilingual + reasoning abilities
22
  > and native HuggingFace ecosystem support (vLLM, llama.cpp, transformers).
@@ -41,8 +42,9 @@ Adds all VLA tokens (video, 3D pose, SNAC audio) on top of the Qwen3 base tokeni
41
  | SNAC Level 1 even | `<snac_132362>` – `<snac_136457>` | 4,096 | 25 Hz fine audio (even frames) |
42
  | SNAC Level 1 odd | `<snac_144650>` – `<snac_148745>` | 4,096 | 25 Hz fine audio (odd frames) |
43
  | SNAC wrappers | `<snac>`, `</snac>` | 2 | Block delimiters |
 
44
 
45
- **Total VLA tokens added: 106,228**
46
 
47
  ---
48
 
@@ -60,7 +62,7 @@ Adds all VLA tokens (video, 3D pose, SNAC audio) on top of the Qwen3 base tokeni
60
  from transformers import AutoTokenizer
61
 
62
  tok = AutoTokenizer.from_pretrained("EmpathicRobotics/tokenizer-vla-qwen3")
63
- print(len(tok)) # 257897
64
 
65
  # All VLA tokens are atomic
66
  print(tok.encode("<seed2_1137>", add_special_tokens=False)) # single ID
@@ -113,7 +115,7 @@ tok.save_pretrained("tokenizer-vla-qwen3")
113
  # vocab size: 257,897
114
  ```
115
 
116
- Full script: `tools/build_tokenizers.py` in the
117
  [finevideo-vla](https://github.com/TieuDaoChanNhan/finevideo-vla) repo.
118
 
119
  ---
 
14
  # VLA Tokenizer — Qwen3
15
 
16
  Extended Qwen3 tokenizer for the **FineVideo-VLA** multimodal dataset.
17
+ Adds all VLA tokens (video, 3D pose, SNAC audio, caption/speech wrappers) on
18
+ top of the Qwen3 base tokenizer.
19
 
20
+ **Vocab size: 257,901** (~151,669 Qwen3 base + 106,232 VLA tokens)
21
 
22
  > **Why Qwen3?** The Qwen3 family has strong multilingual + reasoning abilities
23
  > and native HuggingFace ecosystem support (vLLM, llama.cpp, transformers).
 
42
  | SNAC Level 1 even | `<snac_132362>` – `<snac_136457>` | 4,096 | 25 Hz fine audio (even frames) |
43
  | SNAC Level 1 odd | `<snac_144650>` – `<snac_148745>` | 4,096 | 25 Hz fine audio (odd frames) |
44
  | SNAC wrappers | `<snac>`, `</snac>` | 2 | Block delimiters |
45
+ | Caption/speech wrappers | `<caption>`, `</caption>`, `<speech>`, `</speech>` | 4 | Inline caption/dialogue interleaving |
46
 
47
+ **Total VLA tokens added: 106,232**
48
 
49
  ---
50
 
 
62
  from transformers import AutoTokenizer
63
 
64
  tok = AutoTokenizer.from_pretrained("EmpathicRobotics/tokenizer-vla-qwen3")
65
+ print(len(tok)) # 257901
66
 
67
  # All VLA tokens are atomic
68
  print(tok.encode("<seed2_1137>", add_special_tokens=False)) # single ID
 
115
  # vocab size: 257,897
116
  ```
117
 
118
+ Full script: `tools/tokenizer/build_tokenizers.py` in the
119
  [finevideo-vla](https://github.com/TieuDaoChanNhan/finevideo-vla) repo.
120
 
121
  ---
tokenizer.json CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6da92a63580d67a15989711c8a0817314e64df599610605dd3f8e3f518f84956
3
- size 31654989
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d97a3c88b39e2dffeee8d470d767d48e2777c0d1395aaeee32b3709cb35e68a5
3
+ size 31655733