Upload config.yaml with huggingface_hub
Browse files- config.yaml +51 -0
config.yaml
ADDED
|
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
+
# Config for super vocab built from tokenizers: flexitok/bpe_script_Arab_16000, flexitok/bpe_script_CmJp_16000, flexitok/bpe_ltr_ell_Grek_8000_v2, flexitok/bpe_ltr_fw_edu_32000_v2, flexitok/bpe_ltr_hun_Latn_8000_v2, flexitok/bpe_ltr_rus_Cyrl_16000_v2, flexitok/bpe_ltr_tur_Latn_8000_v2, flexitok/bpe_script_Germ_32000, flexitok/bpe_script_Roma_32000, flexitok/bpe_script_SEAS_16000, flexitok/bpe_script_Slav_16000
|
| 3 |
+
## Training superset tokenizer with individual tokenizers trained on Fineweb-2-hq
|
| 4 |
+
# use with apps/main/configs/flexitok/llama_1b_base.yaml
|
| 5 |
+
##TODO: add router details, make sure name is correct
|
| 6 |
+
model:
|
| 7 |
+
vocab_size: 163711
|
| 8 |
+
name: script_1
|
| 9 |
+
dump_dir: /fsx/craffel/lingua_logs/script_1
|
| 10 |
+
data:
|
| 11 |
+
tokenizer:
|
| 12 |
+
name: supertokenizer
|
| 13 |
+
seed: 42
|
| 14 |
+
superset_code_name: script_1
|
| 15 |
+
n_words: 163711
|
| 16 |
+
tokenizers:
|
| 17 |
+
- name: huggingface
|
| 18 |
+
path: flexitok/bpe_script_Arab_16000
|
| 19 |
+
load_supermapping: true
|
| 20 |
+
- name: huggingface
|
| 21 |
+
path: flexitok/bpe_script_CmJp_16000
|
| 22 |
+
load_supermapping: true
|
| 23 |
+
- name: huggingface
|
| 24 |
+
path: flexitok/bpe_ltr_ell_Grek_8000_v2
|
| 25 |
+
load_supermapping: true
|
| 26 |
+
- name: huggingface
|
| 27 |
+
path: flexitok/bpe_ltr_fw_edu_32000_v2
|
| 28 |
+
load_supermapping: true
|
| 29 |
+
- name: huggingface
|
| 30 |
+
path: flexitok/bpe_ltr_hun_Latn_8000_v2
|
| 31 |
+
load_supermapping: true
|
| 32 |
+
- name: huggingface
|
| 33 |
+
path: flexitok/bpe_ltr_rus_Cyrl_16000_v2
|
| 34 |
+
load_supermapping: true
|
| 35 |
+
- name: huggingface
|
| 36 |
+
path: flexitok/bpe_ltr_tur_Latn_8000_v2
|
| 37 |
+
load_supermapping: true
|
| 38 |
+
- name: huggingface
|
| 39 |
+
path: flexitok/bpe_script_Germ_32000
|
| 40 |
+
load_supermapping: true
|
| 41 |
+
- name: huggingface
|
| 42 |
+
path: flexitok/bpe_script_Roma_32000
|
| 43 |
+
load_supermapping: true
|
| 44 |
+
- name: huggingface
|
| 45 |
+
path: flexitok/bpe_script_SEAS_16000
|
| 46 |
+
load_supermapping: true
|
| 47 |
+
- name: huggingface
|
| 48 |
+
path: flexitok/bpe_script_Slav_16000
|
| 49 |
+
load_supermapping: true
|
| 50 |
+
checkpoint:
|
| 51 |
+
path: /fsx/craffel/lingua_logs/checkpoints/script_1
|