Upload ede-xlm-roberta-base artifacts

README.md

@@ -0,0 +1,31 @@
+
+---
+language:
+- vi
+- en
+license: apache-2.0
+tags:
+- xlm-roberta
+- retrieval
+- splade
+- wechsel
+- vietnamese
+- information-retrieval
+---
+
+# ede-xlm-roberta-base
+
+WECHSEL-initialized XLM-RoBERTa-base encoder for Vietnamese dense/sparse retrieval.
+
+## Included files
+
+- align.pt
+- vocab.json
+- tokenizer_config.json
+- special_tokens.json
+- model_config.json
+- wechsel_embeddings.npy
+
+## Notes
+
+This repository stores tokenizer and alignment artifacts from the WECHSEL pipeline.

align.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:912aa829f15aa5ae9e549499ccc1d84f0e38fce0f61cb81f21bdb80329788c5b
+size 718996371

model_config.json

@@ -0,0 +1,9 @@
+{
+  "architecture": "WECHSEL-XLM-R-SPLADE",
+  "backbone": "xlm-roberta-base",
+  "vocab_size": 60996,
+  "d_model": 768,
+  "max_length": 128,
+  "splade_l1": 0.001,
+  "freeze_layers": 8
+}

special_tokens.json

@@ -0,0 +1,7 @@
+{
+  "PAD_ID": 0,
+  "UNK_ID": 1,
+  "CLS_ID": 2,
+  "SEP_ID": 3,
+  "MASK_ID": 4
+}

tokenizer_config.json

@@ -0,0 +1,12 @@
+{
+  "tokenizer_class": "MorphemeTokenizer",
+  "backbone": "xlm-roberta-base",
+  "vocab_size": 60996,
+  "max_length": 128,
+  "min_freq_word": 2,
+  "min_freq_morph": 3,
+  "wechsel_k": 10,
+  "wechsel_tau": 0.1,
+  "bidict_source": "NIRVLab/rhade-vietnamese-mt",
+  "bidict_anchor_weight": 0.3
+}

wechsel_embeddings.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:610c4a4d3db8c979b709163976264bc648aafd2f8821df6ce236043bd0fba19f
+size 187379840