Multilingual embedding base v1.0 (bge-m3 compatible)

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

1_Pooling/config.json

@@ -0,0 +1,10 @@
+{
+ "word_embedding_dimension": 768,
+ "pooling_mode_cls_token": false,
+ "pooling_mode_mean_tokens": true,
+ "pooling_mode_max_tokens": false,
+ "pooling_mode_mean_sqrt_len_tokens": false,
+ "pooling_mode_weightedmean_tokens": false,
+ "pooling_mode_lasttoken": false,
+ "include_prompt": true
+}

2_Dense/config.json

@@ -0,0 +1,6 @@
+{
+ "in_features": 768,
+ "out_features": 1024,
+ "bias": false,
+ "activation_function": "torch.nn.modules.linear.Identity"
+}

2_Dense/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:54079ba3cc4f5a530208442dc5edc992959a0302bfc029d823cb2ecbc400fbc1
+size 3145816

README.md

@@ -0,0 +1,147 @@
+---
+license: apache-2.0
+language:
+- multilingual
+- en
+- de
+- fr
+- es
+- pt
+- it
+- nl
+- pl
+- ru
+- uk
+- cs
+- ro
+- bg
+- sv
+- da
+- 'no'
+- fi
+- hu
+- el
+- tr
+- ar
+- he
+- fa
+- ur
+- hi
+- bn
+- ta
+- te
+- mr
+- zh
+- ja
+- ko
+- vi
+- th
+- id
+- ms
+- sw
+- af
+- am
+- hy
+- az
+- eu
+- ka
+- kk
+- km
+library_name: sentence-transformers
+pipeline_tag: sentence-similarity
+base_model: jhu-clsp/mmBERT-base
+datasets:
+- HuggingFaceFW/fineweb-2
+- HuggingFaceFW/fineweb
+tags:
+- sentence-transformers
+- feature-extraction
+- sentence-similarity
+- embeddings
+- retrieval
+- multilingual
+- bge-m3
+- distillation
+- modernbert
+- mmbert
+- onnx
+- transformers.js
+---
+
+# Multilingual Embedding (base, 308M) - bge-m3 compatible
+
+A multilingual sentence and passage embedding model distilled from [BAAI/bge-m3](https://huggingface.co/BAAI/bge-m3) into
+[mmBERT-base](https://huggingface.co/jhu-clsp/mmBERT-base). It outputs 1024-dimensional vectors **in bge-m3's dense vector
+space**, so you can:
+
+- use it on its own for search, RAG, clustering and similarity in about 90 languages, 2x smaller than bge-m3;
+- or **query an existing bge-m3 index** with it: embed queries with this model and keep the passages you already embedded
+  with bge-m3 (see the "→ bge-m3 index" row below). That gives cheap query encoding on CPU or in the browser.
+
+A smaller, faster version is available as [Horizon-Labs/multilingual-embedding-small](https://huggingface.co/Horizon-Labs/multilingual-embedding-small). Apache-2.0. ONNX files for CPU and the browser (transformers.js) are included.
+
+- Vectors are L2-normalised. Use cosine similarity (= dot product). No query/passage prefixes are needed.
+- `onnx/model_quantized.onnx` (int8 embeddings, 642 MB; cosine >= 0.99997 to the fp32 model on test sentences) outputs the final normalised embedding as `sentence_embedding`.
+
+## Usage
+
+```python
+from sentence_transformers import SentenceTransformer
+
+model = SentenceTransformer("Horizon-Labs/multilingual-embedding-base")
+q = model.encode(["How tall is the Eiffel Tower?"])
+p = model.encode(["La tour Eiffel mesure 330 mètres.", "The Statue of Liberty is 93 metres tall."])
+print(q @ p.T)   # higher = more similar
+
+# Querying an existing bge-m3 index: passages embedded earlier with BAAI/bge-m3 (dense, normalised)
+# scores = model.encode(queries) @ bge_m3_passage_vectors.T
+```
+
+transformers.js:
+
+```js
+import { AutoTokenizer, AutoModel } from "@huggingface/transformers";
+const tok = await AutoTokenizer.from_pretrained("Horizon-Labs/multilingual-embedding-base");
+const model = await AutoModel.from_pretrained("Horizon-Labs/multilingual-embedding-base", { dtype: "q8" });
+const { sentence_embedding } = await model(await tok(["Wie hoch ist der Eiffelturm?"], { padding: true, truncation: true }));
+```
+
+## Evaluation
+
+nDCG@10 for ranking each query's candidates by cosine similarity, on public reranking benchmarks (MTEB versions; used
+only for evaluation). MIRACL: 60 queries per language with 100 candidates each. Wikipedia: 60 queries per language with 9
+candidates each. "Other" = ESCI (es, jp, us), RuBQ, T2Reranking, mMARCO-ja and AskUbuntu. Same script for every model
+(`code/`); multilingual-e5 is run with its "query: "/"passage: " prefixes.
+
+| model | MIRACL (18 languages) | Wikipedia (16 languages) | other (6 sets) | mean | note |
+|---|---|---|---|---|---|
+| **this model** (308M) | 0.753 | 0.934 | 0.777 | 0.821 | queries and passages embedded by this model |
+| **this model** → bge-m3 index (308M) | 0.742 | 0.934 | 0.774 | 0.817 | queries by this model, passages by bge-m3 |
+| [Horizon-Labs/multilingual-embedding-small](https://huggingface.co/Horizon-Labs/multilingual-embedding-small) (141M) | 0.750 | 0.935 | 0.768 | 0.818 |  |
+| [BAAI/bge-m3](https://huggingface.co/BAAI/bge-m3) (568M) | 0.797 | 0.927 | 0.788 | 0.837 | the teacher |
+| [intfloat/multilingual-e5-small](https://huggingface.co/intfloat/multilingual-e5-small) (118M) | 0.721 | 0.915 | 0.776 | 0.804 |  |
+| [intfloat/multilingual-e5-base](https://huggingface.co/intfloat/multilingual-e5-base) (278M) | 0.736 | 0.918 | 0.775 | 0.810 |  |
+| [sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2](https://huggingface.co/sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2) (118M) | 0.387 | 0.837 | 0.687 | 0.637 |  |
+| [sentence-transformers/all-MiniLM-L6-v2](https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2) (23M) | 0.172 | 0.720 | 0.621 | 0.504 | English |
+
+- The table shows the released checkpoint. Means over two training seeds: small own .817 / against a bge-m3 index .818; base own .822 / against a bge-m3 index .816.
+- Models ahead of this one (own embeddings): MIRACL: bge-m3; Wikipedia: none;
+  other: bge-m3.
+- This is a reranking-style test over given candidate lists, not a full-corpus retrieval benchmark.
+
+## Training
+
+- **Distillation**: the student (mmBERT-base + mean pooling + a linear layer to 1024 dimensions) learns to reproduce bge-m3's
+  normalised dense embeddings: cosine + squared-error loss, plus a loss on the in-batch similarity matrix that keeps the
+  relative geometry.
+- **Text**: about 4.65 million texts: snippets of 1-8 sentences and short spans from [FineWeb-2](https://huggingface.co/datasets/HuggingFaceFW/fineweb-2)
+  and [FineWeb](https://huggingface.co/datasets/HuggingFaceFW/fineweb) (ODC-BY) in 92 languages, plus about 850k search queries
+  written by Qwen3.8-27B (Apache-2.0) for web passages in 47 languages.
+- Code: `code/` in this repository.
+
+## Limitations
+
+- The student approximates bge-m3 and is weaker than it. The gap is larger for long passages, rare languages and
+  specialised domains.
+- Only bge-m3's **dense** vectors are reproduced, not its sparse or multi-vector (ColBERT) outputs.
+- Max length 512 tokens (trained with 256). Split long documents into passages.

code/emb/build_release.py

@@ -0,0 +1,67 @@
+"""Package an embedding student as a sentence-transformers model + ONNX (sentence_embedding output) + int8 check.
+python emb/build_release.py SRC_DIR BASE_HUB_ID OUT_DIR
+SRC_DIR: emb/train_emb.py output (encoder + projection.pt). OUT_DIR: Transformer weights at the root, 1_Pooling (mean),
+2_Dense (hidden -> 1024, no bias), 3_Normalize; onnx/model.onnx + onnx/model_quantized.onnx (inputs input_ids,
+attention_mask; output sentence_embedding = normalised mean-pooled projection) and onnx_sweep.json (cosine fp32 vs int8).
+"""
+import json, os, shutil, sys
+import numpy as np, torch
+from huggingface_hub import hf_hub_download
+from safetensors.torch import save_file
+from transformers import AutoModel, AutoTokenizer
+
+src, base, out = sys.argv[1:4]
+os.makedirs(out, exist_ok=True)
+for f in ["config.json", "model.safetensors", "tokenizer.json", "val_metrics.json", "train_log.json"]:
+    if os.path.exists(f"{src}/{f}"):
+        shutil.copy(f"{src}/{f}", f"{out}/{f}")
+for f in ["tokenizer_config.json", "special_tokens_map.json"]:
+    shutil.copy(hf_hub_download(base, f), f"{out}/{f}")
+P = torch.load(f"{src}/projection.pt", map_location="cpu").float()   # (hidden, 1024)
+hid = P.shape[0]
+json.dump([{"idx": 0, "name": "0", "path": "", "type": "sentence_transformers.models.Transformer"},
+           {"idx": 1, "name": "1", "path": "1_Pooling", "type": "sentence_transformers.models.Pooling"},
+           {"idx": 2, "name": "2", "path": "2_Dense", "type": "sentence_transformers.models.Dense"},
+           {"idx": 3, "name": "3", "path": "3_Normalize", "type": "sentence_transformers.models.Normalize"}], open(f"{out}/modules.json", "w"), indent=1)
+json.dump({"max_seq_length": 512, "do_lower_case": False}, open(f"{out}/sentence_bert_config.json", "w"))
+json.dump({"similarity_fn_name": "cosine", "prompts": {}, "default_prompt_name": None}, open(f"{out}/config_sentence_transformers.json", "w"))
+os.makedirs(f"{out}/1_Pooling", exist_ok=True); os.makedirs(f"{out}/2_Dense", exist_ok=True); os.makedirs(f"{out}/3_Normalize", exist_ok=True)
+json.dump({"word_embedding_dimension": hid, "pooling_mode_cls_token": False, "pooling_mode_mean_tokens": True, "pooling_mode_max_tokens": False,
+           "pooling_mode_mean_sqrt_len_tokens": False, "pooling_mode_weightedmean_tokens": False, "pooling_mode_lasttoken": False,
+           "include_prompt": True}, open(f"{out}/1_Pooling/config.json", "w"), indent=1)
+json.dump({"in_features": hid, "out_features": 1024, "bias": False, "activation_function": "torch.nn.modules.linear.Identity"},
+          open(f"{out}/2_Dense/config.json", "w"), indent=1)
+save_file({"linear.weight": P.T.contiguous()}, f"{out}/2_Dense/model.safetensors")
+
+tok = AutoTokenizer.from_pretrained(out); enc_m = AutoModel.from_pretrained(out, attn_implementation="eager").eval()
+
+
+class Wrap(torch.nn.Module):
+    def __init__(self, m, P):
+        super().__init__(); self.m = m; self.P = torch.nn.Parameter(P, requires_grad=False)
+
+    def forward(self, input_ids, attention_mask):
+        h = self.m(input_ids=input_ids, attention_mask=attention_mask).last_hidden_state
+        a = attention_mask.unsqueeze(-1).to(h.dtype); v = (h * a).sum(1) / a.sum(1)
+        return torch.nn.functional.normalize(v @ self.P, dim=-1)
+
+
+texts = ["How tall is the Eiffel Tower?", "La tour Eiffel mesure 330 mètres.", "电动汽车电池能用多久", "Der Eiffelturm ist mit 330 Metern das höchste Bauwerk von Paris. " * 8]
+e = tok(texts, return_tensors="pt", padding=True)
+os.makedirs(f"{out}/onnx", exist_ok=True); w = Wrap(enc_m, P).eval()
+with torch.no_grad():
+    ref = w(e["input_ids"], e["attention_mask"]).numpy()
+    torch.onnx.export(w, (e["input_ids"], e["attention_mask"]), f"{out}/onnx/model.onnx", input_names=["input_ids", "attention_mask"],
+                      output_names=["sentence_embedding"], dynamic_axes={"input_ids": {0: "batch", 1: "seq"}, "attention_mask": {0: "batch", 1: "seq"},
+                                                                          "sentence_embedding": {0: "batch"}}, opset_version=17, dynamo=False)
+import onnxruntime as ort
+from onnxruntime.quantization import quantize_dynamic, QuantType
+quantize_dynamic(f"{out}/onnx/model.onnx", f"{out}/onnx/model_quantized.onnx", weight_type=QuantType.QInt8, op_types_to_quantize=["Gather"])
+feeds = {k: e[k].numpy() for k in ["input_ids", "attention_mask"]}
+r = {}
+for f in ["model.onnx", "model_quantized.onnx"]:
+    o = ort.InferenceSession(f"{out}/onnx/{f}", providers=["CPUExecutionProvider"]).run(None, feeds)[0]
+    r[f] = dict(cos_to_torch=float((o * ref).sum(-1).min()), mb=os.path.getsize(f"{out}/onnx/{f}") / 1e6)
+json.dump(dict(gather_only=dict(min_cos=r["model_quantized.onnx"]["cos_to_torch"], mb=r["model_quantized.onnx"]["mb"]), fp32=r["model.onnx"], n=len(texts)),
+          open(f"{out}/onnx_sweep.json", "w"), indent=1)
+print(r, flush=True)

code/emb/build_texts.py

@@ -0,0 +1,71 @@
+"""Text pool for embedding distillation: FineWeb-2 / FineWeb (ODC-BY) snippets of 1-8 consecutive sentences (20-2000 chars;
+a third are single sentences or short spans), ~90 languages.   python emb/build_texts.py OUT.parquet [PER_LANG]   (CPU job)
+Columns: pid, text, lang, url. English gets 4x PER_LANG. Different random seed and documents than rerank/build_passages.py.
+"""
+import json, os, random, re, sys, time
+from multiprocessing import Pool
+
+OUT = sys.argv[1]
+PER = int(sys.argv[2]) if len(sys.argv) > 2 else 8000
+FW = json.load(open(os.path.join(os.path.dirname(os.path.abspath(__file__)), "..", "lid", "fw_files.json")))
+LANGS2 = ("afr_Latn als_Latn amh_Ethi hye_Armn azj_Latn eus_Latn bel_Cyrl bos_Latn cym_Latn epo_Latn glg_Latn guj_Gujr hau_Latn "
+          "isl_Latn gle_Latn jav_Latn kan_Knda kat_Geor kaz_Cyrl khm_Khmr kir_Cyrl lao_Laoo mal_Mlym mkd_Cyrl mlt_Latn khk_Cyrl mya_Mymr "
+          "npi_Deva ory_Orya pan_Guru pbt_Arab sin_Sinh som_Latn sun_Latn tgk_Cyrl tat_Cyrl uig_Arab uzn_Latn xho_Latn zul_Latn "
+          "ltz_Latn fao_Latn snd_Arab ckb_Arab kin_Latn").split()
+LANGS = ("arb_Arab ben_Beng deu_Latn spa_Latn fas_Arab fin_Latn fra_Latn hin_Deva ind_Latn jpn_Jpan kor_Hang rus_Cyrl swh_Latn "
+         "tel_Telu tha_Thai yor_Latn cmn_Hani bul_Cyrl ces_Latn dan_Latn ita_Latn nld_Latn nob_Latn por_Latn ron_Latn srp_Cyrl "
+         "swe_Latn pol_Latn tur_Latn ukr_Cyrl vie_Latn heb_Hebr ell_Grek hun_Latn zsm_Latn fil_Latn urd_Arab tam_Taml mar_Deva "
+         "slk_Latn hrv_Latn cat_Latn lit_Latn ekk_Latn lvs_Latn slv_Latn").split()
+SPLIT = re.compile(r"(?<=[.!?。！？।؟።])\s+|\n+")
+
+
+def collect(args):
+    lang, repo, files, per = args
+    import pyarrow.parquet as pq
+    from huggingface_hub import HfFileSystem
+    rng = random.Random(1007 + sum(map(ord, lang))); fs = HfFileSystem(); rows = []
+
+    def retry(fn):
+        for k in range(6):
+            try:
+                return fn()
+            except Exception:
+                if k == 5:
+                    raise
+                time.sleep(20 * (k + 1))
+    try:
+        for f in files:
+            pf = retry(lambda: pq.ParquetFile(fs.open(f"datasets/{repo}/{f}")))
+            for g in range(pf.num_row_groups - 1, -1, -1):   # last row groups first: other documents than the passage pool
+                tb = retry(lambda: pf.read_row_group(g, columns=["text", "url"]))
+                for text, url in zip(tb.column("text").to_pylist(), tb.column("url").to_pylist()):
+                    sents = [s.strip() for s in SPLIT.split(text or "") if len(s.strip()) >= 10]
+                    if len(sents) < 2:
+                        continue
+                    k = rng.choice([1, 1, 2, 3, 4, 6, 8]); i = rng.randrange(max(1, len(sents) - k + 1))
+                    t = " ".join(sents[i:i + k])
+                    if rng.random() < 0.15 and len(t) > 60:   # short span (query-like)
+                        st = rng.randrange(0, len(t) - 40); t = t[st:st + rng.randint(20, 60)]
+                    if 20 <= len(t) <= 2000:
+                        rows.append((t, lang, url))
+                    if len(rows) >= per:
+                        break
+                if len(rows) >= per:
+                    break
+            if len(rows) >= per:
+                break
+    except Exception as e:
+        print(lang, "failed", e, flush=True)
+    print(lang, len(rows), flush=True)
+    return rows
+
+
+if __name__ == "__main__":
+    import pandas as pd
+    jobs = [("eng_Latn", "HuggingFaceFW/fineweb", FW["fineweb_eng"], PER * 4)] + [(l, "HuggingFaceFW/fineweb-2", FW["fineweb2"][l], PER) for l in LANGS + [x for x in LANGS2 if x in FW["fineweb2"]]]
+    with Pool(int(os.environ.get("NPROC", 14))) as p:
+        res = p.map(collect, jobs)
+    df = pd.DataFrame([r for rs in res for r in rs], columns=["text", "lang", "url"]).drop_duplicates("text").reset_index(drop=True)
+    df.insert(0, "pid", range(len(df)))
+    df.to_parquet(OUT); print("passages", len(df), df.lang.nunique(), "langs", flush=True)
+    os._exit(0)

code/emb/embed_teacher.py

@@ -0,0 +1,29 @@
+"""bge-m3 (MIT) dense embeddings as distillation targets.   python emb/embed_teacher.py OUT_DIR SRC [SRC ...]
+SRC = parquet path[:column] (default column "text"). Writes OUT_DIR/texts.parquet (text, src) and OUT_DIR/emb.npy
+(float16, n x 1024, CLS pooling, L2-normalised, max 512 tokens) in the same order. Duplicated texts are embedded once.
+"""
+import os, sys
+import numpy as np, pandas as pd, torch
+from transformers import AutoModel, AutoTokenizer
+
+out = sys.argv[1]; os.makedirs(out, exist_ok=True)
+parts = []
+for spec in sys.argv[2:]:
+    path, col = (spec.rsplit(":", 1) if spec.endswith((":text", ":query")) else (spec, "text"))
+    parts.append(pd.DataFrame({"text": pd.read_parquet(path, columns=[col])[col].astype(str), "src": os.path.basename(path) + ":" + col}))
+df = pd.concat(parts, ignore_index=True).drop_duplicates("text").reset_index(drop=True)
+sh, ns = int(os.environ.get("SHARD", 0)), int(os.environ.get("NSHARD", 1))   # optional sharding (rows sh::ns)
+df = df.iloc[sh::ns].reset_index(drop=True)
+print("texts", len(df), df.src.value_counts().to_dict(), flush=True)
+tok = AutoTokenizer.from_pretrained("BAAI/bge-m3"); m = AutoModel.from_pretrained("BAAI/bge-m3", torch_dtype=torch.float16).cuda().eval()
+E = np.lib.format.open_memmap(f"{out}/emb.npy", mode="w+", dtype=np.float16, shape=(len(df), 1024))
+texts = df.text.tolist(); order = np.argsort([len(t) for t in texts]); bs = 256
+with torch.no_grad():
+    for s in range(0, len(texts), bs):
+        b = order[s:s + bs]
+        e = tok([texts[i] for i in b], truncation=True, max_length=512, padding=True, return_tensors="pt").to("cuda")
+        h = m(**e).last_hidden_state[:, 0]
+        E[np.sort(b)] = torch.nn.functional.normalize(h.float(), dim=-1).half().cpu().numpy()[np.argsort(b)]
+        if s % (bs * 400) == 0:
+            print(s, flush=True)
+E.flush(); df.to_parquet(f"{out}/texts.parquet"); print("done", flush=True)

code/emb/eval_emb.py

@@ -0,0 +1,78 @@
+"""Evaluate embedding models by cosine reranking on rerank/evals (nDCG@10, same sets/metric as rerank/eval_rerank.py).
+python emb/eval_emb.py --models A,B --evals rerank/evals --out R.json
+Specs: cls:<id> (CLS pooling, e.g. BAAI/bge-m3) | mean:<id> (mean pooling) | e5:<id> (mean pooling + "query: "/"passage: "
+prefixes) | ours:<dir> (our student: mean pooling + dir/projection.pt) | asym:<dir> (queries by our student, documents by
+bge-m3 - querying an existing bge-m3 index). All embeddings L2-normalised; documents truncated to 512 tokens.
+"""
+import argparse, glob, json, math, os, sys, time
+import numpy as np, pandas as pd, torch
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+ap = argparse.ArgumentParser()
+ap.add_argument("--models", required=True); ap.add_argument("--evals", required=True); ap.add_argument("--out", required=True)
+ap.add_argument("--bs", type=int, default=128)
+args = ap.parse_args()
+EV = {os.path.basename(p)[:-8]: pd.read_parquet(p) for p in sorted(glob.glob(f"{args.evals}/*.parquet"))}
+
+
+def ndcg10(rels, scores):
+    order = np.argsort(-np.asarray(scores), kind="stable")[:10]
+    dcg = sum((2 ** rels[i] - 1) / math.log2(k + 2) for k, i in enumerate(order))
+    idcg = sum((2 ** r - 1) / math.log2(k + 2) for k, r in enumerate(sorted(rels, reverse=True)[:10]))
+    return dcg / idcg if idcg > 0 else 0.0
+
+
+class Enc:
+    def __init__(self, spec):
+        from transformers import AutoModel, AutoTokenizer
+        self.kind, mid = spec.split(":", 1)
+        self.tok = AutoTokenizer.from_pretrained(mid); self.m = AutoModel.from_pretrained(mid, torch_dtype=torch.float16).cuda().eval()
+        self.proj = None
+        if self.kind == "ours":
+            self.proj = torch.load(f"{mid}/projection.pt", map_location="cuda").half()
+
+    @torch.no_grad()
+    def __call__(self, texts, is_query):
+        if self.kind == "e5":
+            texts = [("query: " if is_query else "passage: ") + t for t in texts]
+        out = np.zeros((len(texts), 0), dtype=np.float32); chunks = [None] * len(texts); order = np.argsort([len(t) for t in texts])
+        for s in range(0, len(texts), args.bs):
+            b = order[s:s + args.bs]
+            e = self.tok([texts[i] for i in b], truncation=True, max_length=64 if is_query else 512, padding=True, return_tensors="pt").to("cuda")
+            h = self.m(input_ids=e["input_ids"], attention_mask=e["attention_mask"]).last_hidden_state
+            if self.kind == "cls":
+                v = h[:, 0]
+            else:
+                a = e["attention_mask"].unsqueeze(-1).to(h.dtype); v = (h * a).sum(1) / a.sum(1)
+            if self.proj is not None:
+                v = v @ self.proj
+            v = torch.nn.functional.normalize(v.float(), dim=-1).cpu().numpy()
+            for j, i in enumerate(b):
+                chunks[i] = v[j]
+        return np.stack(chunks)
+
+
+res = json.load(open(args.out)) if os.path.exists(args.out) else {}
+bge = None
+for spec in args.models.split(","):
+    if spec.startswith("asym:"):
+        bge = bge or Enc("cls:BAAI/bge-m3"); qenc, denc = Enc("ours:" + spec[5:]), bge
+    else:
+        qenc = denc = Enc(spec)
+    r, t0, n = {}, time.time(), 0
+    for name, df in EV.items():
+        Q = qenc(df["query"].tolist(), True)
+        docs = [d for ds in df.docs for d in ds]; D = denc(docs, False); n += len(docs)
+        k = 0; nd = []
+        for qi, rels in enumerate(df.rels):
+            s = D[k:k + len(rels)] @ Q[qi]; k += len(rels); nd.append(ndcg10(list(rels), s))
+        r[name] = dict(n=len(df), ndcg10=float(np.mean(nd)))
+    fam = lambda p: [v["ndcg10"] for k2, v in r.items() if k2.startswith(p)]
+    r["_miracl"], r["_wiki"] = float(np.mean(fam("miracl_"))), float(np.mean(fam("wiki_")))
+    r["_other"] = float(np.mean([v["ndcg10"] for k2, v in r.items() if not k2.startswith(("miracl_", "wiki_", "_"))]))
+    r["_all"] = float(np.mean([r["_miracl"], r["_wiki"], r["_other"]])); r["_docs_per_s"] = n / (time.time() - t0)
+    print(spec, {k2: round(v, 4) for k2, v in r.items() if k2.startswith("_")}, flush=True)
+    res[spec] = r; json.dump(res, open(args.out, "w"), indent=1)
+    if not spec.startswith("asym:"):
+        del qenc, denc
+    torch.cuda.empty_cache()

code/emb/train_emb.py

@@ -0,0 +1,78 @@
+"""Distil BAAI/bge-m3 dense embeddings into a smaller encoder (same 1024-d vector space).
+python emb/train_emb.py --base jhu-clsp/mmBERT-small --teacher DIR[,DIR] --out OUT
+Student = encoder + mean pooling + linear projection (hidden -> 1024, saved as OUT/projection.pt), L2-normalised.
+Loss = (1 - cos(student, teacher)) + mse_w * ||student - teacher||^2 + sim_w * in-batch similarity-matrix MSE (keeps the
+relative geometry). Val = 0.5% of texts: mean cosine to the teacher and top-1 in-batch retrieval agreement.
+"""
+import argparse, json, os, random, time
+import numpy as np, pandas as pd, torch
+import torch.nn.functional as F
+from transformers import AutoModel, AutoTokenizer, get_cosine_schedule_with_warmup
+
+ap = argparse.ArgumentParser()
+ap.add_argument("--base", default="jhu-clsp/mmBERT-small"); ap.add_argument("--teacher", required=True); ap.add_argument("--out", required=True)
+ap.add_argument("--max_len", type=int, default=256); ap.add_argument("--bs", type=int, default=256); ap.add_argument("--lr", type=float, default=1e-4)
+ap.add_argument("--epochs", type=float, default=2.0); ap.add_argument("--warmup", type=float, default=0.03)
+ap.add_argument("--mse_w", type=float, default=1.0); ap.add_argument("--sim_w", type=float, default=1.0)
+ap.add_argument("--seed", type=int, default=0); ap.add_argument("--eval_every", type=int, default=2000)
+args = ap.parse_args()
+random.seed(args.seed); np.random.seed(args.seed); torch.manual_seed(args.seed)
+os.makedirs(args.out, exist_ok=True)
+texts, embs = [], []
+for d in args.teacher.split(","):
+    texts += pd.read_parquet(f"{d}/texts.parquet").text.tolist(); embs.append(np.load(f"{d}/emb.npy", mmap_mode="r"))
+E = np.concatenate([np.asarray(e) for e in embs]).astype(np.float16)
+n = len(texts); rng = np.random.RandomState(0); isval = rng.rand(n) < 0.005
+tr_idx, va_idx = np.where(~isval)[0], np.where(isval)[0][:4096]
+print("texts", n, "train", len(tr_idx), "val", len(va_idx), flush=True)
+tok = AutoTokenizer.from_pretrained(args.base); enc = AutoModel.from_pretrained(args.base).cuda()
+proj = torch.nn.Linear(enc.config.hidden_size, 1024, bias=False).cuda()
+params = list(enc.parameters()) + list(proj.parameters())
+
+
+def embed(idx):
+    e = tok([texts[i] for i in idx], truncation=True, max_length=args.max_len, padding=True, return_tensors="pt").to("cuda")
+    with torch.autocast("cuda", dtype=torch.bfloat16):
+        h = enc(input_ids=e["input_ids"], attention_mask=e["attention_mask"]).last_hidden_state
+    a = e["attention_mask"].unsqueeze(-1).float(); v = (h.float() * a).sum(1) / a.sum(1)
+    return F.normalize(proj(v), dim=-1)
+
+
+def loss_fn(s, t):
+    cos = (s * t).sum(-1)
+    return (1 - cos).mean() + args.mse_w * ((s - t) ** 2).sum(-1).mean() + args.sim_w * ((s @ s.T - t @ t.T) ** 2).mean() * 100
+
+
+@torch.no_grad()
+def evaluate():
+    enc.eval(); S = []
+    for s in range(0, len(va_idx), 256):
+        S.append(embed(va_idx[s:s + 256]))
+    enc.train(); S = torch.cat(S); T = torch.tensor(E[va_idx], dtype=torch.float32, device="cuda")
+    top = ((S @ T.T).argmax(1) == torch.arange(len(S), device="cuda")).float().mean().item()
+    return dict(cos=float((S * T).sum(-1).mean().item()), top1=top)
+
+
+steps = int(len(tr_idx) / args.bs * args.epochs)
+opt = torch.optim.AdamW(params, lr=args.lr, weight_decay=0.01, betas=(0.9, 0.98), eps=1e-6, fused=True)
+sch = get_cosine_schedule_with_warmup(opt, int(args.warmup * steps), steps)
+print("steps", steps, flush=True)
+step, best, log, t0 = 0, -1, [], time.time()
+while step < steps:
+    perm = np.random.permutation(tr_idx)
+    for s in range(0, len(perm) - args.bs + 1, args.bs):
+        b = perm[s:s + args.bs]
+        st = embed(b); tt = torch.tensor(E[b], dtype=torch.float32, device="cuda")
+        loss = loss_fn(st, tt); loss.backward(); torch.nn.utils.clip_grad_norm_(params, 1.0)
+        opt.step(); sch.step(); opt.zero_grad(set_to_none=True); step += 1
+        if step % 200 == 0:
+            print(f"step {step}/{steps} loss {loss.item():.4f} {time.time()-t0:.0f}s", flush=True)
+        if step % args.eval_every == 0 or step == steps:
+            r = evaluate(); log.append(dict(step=step, **r)); print("EVAL", step, json.dumps(r), flush=True)
+            if r["cos"] > best:
+                best = r["cos"]; enc.save_pretrained(args.out); tok.save_pretrained(args.out); torch.save(proj.weight.detach().T.contiguous().cpu(), f"{args.out}/projection.pt")
+                json.dump(dict(step=step, **r), open(f"{args.out}/val_metrics.json", "w"), indent=1)
+        if step >= steps:
+            break
+json.dump(dict(args=vars(args), log=log), open(f"{args.out}/train_log.json", "w"), indent=1)
+print("done; best val cos", best)

code/rerank/build_evals.py

@@ -0,0 +1,56 @@
+"""Reranking evaluation sets from MTEB reranking datasets (evaluation only).   python rerank/build_evals.py OUT_DIR
+Each output parquet = one (dataset, language): rows qid, query, docs (candidate texts, in the dataset's candidate order),
+rels (graded relevance per candidate). Queries are sampled (seed 0) up to CAP per set; all candidates of a query are kept.
+Sets: MIRACL (18 languages, dev, CC-BY-SA-4.0), WikipediaRerankingMultilingual (CC-BY-SA-3.0), ESCI (product search,
+Apache-2.0), RuBQ (ru), T2Reranking (zh), VoyageMMarco (ja), AskUbuntuDupQuestions and StackOverflowDupQuestions (en).
+"""
+import os, sys
+import pandas as pd
+from huggingface_hub import HfApi, hf_hub_download
+
+OUT = sys.argv[1]
+os.makedirs(OUT, exist_ok=True)
+api = HfApi()
+SETS = [("mteb/MIRACLReranking", "miracl", 60), ("mteb/WikipediaRerankingMultilingual", "wiki", 60), ("mteb/ESCIReranking", "esci", 150),
+        ("mteb/RuBQReranking", "rubq", 150), ("mteb/T2Reranking", "t2", 150), ("mteb/VoyageMMarcoReranking", "mmarco", 150),
+        ("mteb/AskUbuntuDupQuestions", "askubuntu", 361), ("mteb/stackoverflowdupquestions-reranking", "stackoverflow", 300)]
+
+
+def load(repo, files, part, split):
+    cand = [f for f in files if (f.startswith(part + "/") or f.split("/")[0].endswith("-" + part)) and f.split("/")[-1].startswith(split)]
+    return cand
+
+
+for repo, name, cap in SETS:
+    files = [s.rfilename for s in api.dataset_info(repo).siblings if s.rfilename.endswith(".parquet")]
+    split = "test" if any("/test-" in f for f in files) else "dev"
+    prefixes = sorted({f.split("/")[0].rsplit("-", 1)[0] for f in files if "-" in f.split("/")[0]}) or [""]
+    for pre in prefixes:
+        def part(p):
+            alts = [f"{pre}-{p}/{split}" if pre else f"{p}/{split}"] + ([f"{pre}-qrels/{split}"] if p == "data" else [])
+            for f in files:
+                if any(f.startswith(a) for a in alts):
+                    return pd.read_parquet(hf_hub_download(repo, f, repo_type="dataset"))
+            return None
+        corpus, queries, top = part("corpus"), part("queries"), part("top_ranked")
+        qrels = part("qrels")
+        if qrels is None:
+            qrels = part("data")
+        if corpus is None or queries is None or top is None or qrels is None:
+            print(name, pre, "incomplete; skipped"); continue
+        ctext = dict(zip(corpus["_id"], (corpus.get("title", "").fillna("") + " " + corpus.text).str.strip().str.slice(0, 2000)))
+        qtext = dict(zip(queries["_id"], queries.text))
+        rel = {(q, c): s for q, c, s in zip(qrels["query-id"], qrels["corpus-id"], qrels["score"])}
+        top = top[top["query-id"].isin(qtext)].sample(frac=1.0, random_state=0)
+        rows = []
+        for q, cids in zip(top["query-id"], top["corpus-ids"]):
+            cids = [c for c in cids if c in ctext]
+            rels = [int(rel.get((q, c), 0)) for c in cids]
+            if not any(r > 0 for r in rels) or all(r > 0 for r in rels):
+                continue
+            rows.append((q, qtext[q], [ctext[c] for c in cids], rels))
+            if len(rows) >= cap:
+                break
+        tag = f"{name}_{pre}" if pre else name
+        pd.DataFrame(rows, columns=["qid", "query", "docs", "rels"]).to_parquet(f"{OUT}/{tag}.parquet")
+        print(tag, len(rows), "queries,", sum(len(r[2]) for r in rows), "pairs", flush=True)

code/rerank/build_passages.py

@@ -0,0 +1,65 @@
+"""Passage pool for reranker training: FineWeb-2 / FineWeb (ODC-BY) chunks of 2-6 consecutive sentences (120-1200 chars).
+python rerank/build_passages.py OUT.parquet [PER_LANG]   (CPU job; file list from lid/fw_files.json)
+Columns: pid, text, lang (FineWeb-2 subset code), url. English gets 4x PER_LANG.
+"""
+import json, os, random, re, sys, time
+from multiprocessing import Pool
+
+OUT = sys.argv[1]
+PER = int(sys.argv[2]) if len(sys.argv) > 2 else 8000
+FW = json.load(open(os.path.join(os.path.dirname(os.path.abspath(__file__)), "..", "lid", "fw_files.json")))
+LANGS = ("arb_Arab ben_Beng deu_Latn spa_Latn fas_Arab fin_Latn fra_Latn hin_Deva ind_Latn jpn_Jpan kor_Hang rus_Cyrl swh_Latn "
+         "tel_Telu tha_Thai yor_Latn cmn_Hani bul_Cyrl ces_Latn dan_Latn ita_Latn nld_Latn nob_Latn por_Latn ron_Latn srp_Cyrl "
+         "swe_Latn pol_Latn tur_Latn ukr_Cyrl vie_Latn heb_Hebr ell_Grek hun_Latn zsm_Latn fil_Latn urd_Arab tam_Taml mar_Deva "
+         "slk_Latn hrv_Latn cat_Latn lit_Latn ekk_Latn lvs_Latn slv_Latn").split()
+SPLIT = re.compile(r"(?<=[.!?。！？।؟።])\s+|\n+")
+
+
+def collect(args):
+    lang, repo, files, per = args
+    import pyarrow.parquet as pq
+    from huggingface_hub import HfFileSystem
+    rng = random.Random(7 + sum(map(ord, lang))); fs = HfFileSystem(); rows = []
+
+    def retry(fn):
+        for k in range(6):
+            try:
+                return fn()
+            except Exception:
+                if k == 5:
+                    raise
+                time.sleep(20 * (k + 1))
+    try:
+        for f in files:
+            pf = retry(lambda: pq.ParquetFile(fs.open(f"datasets/{repo}/{f}")))
+            for g in range(pf.num_row_groups):
+                tb = retry(lambda: pf.read_row_group(g, columns=["text", "url"]))
+                for text, url in zip(tb.column("text").to_pylist(), tb.column("url").to_pylist()):
+                    sents = [s.strip() for s in SPLIT.split(text or "") if len(s.strip()) >= 10]
+                    if len(sents) < 2:
+                        continue
+                    k = rng.randint(2, 6); i = rng.randrange(max(1, len(sents) - k + 1))
+                    t = " ".join(sents[i:i + k])
+                    if 120 <= len(t) <= 1200:
+                        rows.append((t, lang, url))
+                    if len(rows) >= per:
+                        break
+                if len(rows) >= per:
+                    break
+            if len(rows) >= per:
+                break
+    except Exception as e:
+        print(lang, "failed", e, flush=True)
+    print(lang, len(rows), flush=True)
+    return rows
+
+
+if __name__ == "__main__":
+    import pandas as pd
+    jobs = [("eng_Latn", "HuggingFaceFW/fineweb", FW["fineweb_eng"], PER * 4)] + [(l, "HuggingFaceFW/fineweb-2", FW["fineweb2"][l], PER) for l in LANGS]
+    with Pool(int(os.environ.get("NPROC", 14))) as p:
+        res = p.map(collect, jobs)
+    df = pd.DataFrame([r for rs in res for r in rs], columns=["text", "lang", "url"]).drop_duplicates("text").reset_index(drop=True)
+    df.insert(0, "pid", range(len(df)))
+    df.to_parquet(OUT); print("passages", len(df), df.lang.nunique(), "langs", flush=True)
+    os._exit(0)

code/rerank/gen_queries.py

@@ -0,0 +1,56 @@
+"""Search queries for passages, Qwen3.8-27B via vLLM (Apache-2.0 outputs).
+IN=passages.parquet SHARD=i NSHARD=n OUT=dir python rerank/gen_queries.py -> $OUT/q_<shard>.jsonl {pid, query, qtype, qlang}
+Per passage: one natural-language question and one short keyword-style query, in the passage's language; for 15% of
+non-English passages one more query in English (cross-lingual search). Queries must be answerable by the passage but
+should not copy long phrases from it.
+"""
+import json, os, random, re, sys
+sys.path.insert(0, os.path.join(os.path.dirname(os.path.abspath(__file__)), "..", "sentiment"))
+
+
+def parse_json(t):
+    t = re.sub(r"^```(?:json)?|```$", "", t.strip(), flags=re.M).strip()
+    s, e = t.find("{"), t.rfind("}")
+    try:
+        return json.loads(t[s: e + 1])
+    except Exception:
+        return None
+
+
+def prompt(text, xl):
+    extra = ('\n- "english": a natural question in English that this passage answers (for cross-lingual search)' if xl else "")
+    return f"""Passage:
+<<<
+{text}
+>>>
+
+Write search queries that a user might type into a search engine and for which this passage is a good result.
+- "question": a natural question in the same language as the passage, answered by the passage
+- "keywords": a short keyword-style query (2-6 words) in the same language as the passage{extra}
+Do not copy long phrases from the passage; use the words a searcher would use. Return only JSON with these keys."""
+
+
+if __name__ == "__main__":
+    import pandas as pd
+    from vllm import LLM, SamplingParams
+    from common import hard_exit
+    df = pd.read_parquet(os.environ["IN"]); sh, ns = int(os.environ.get("SHARD", 0)), int(os.environ.get("NSHARD", 1))
+    df = df.iloc[sh::ns]
+    rng = random.Random(11 + sh)
+    xl = [l != "eng_Latn" and rng.random() < 0.15 for l in df.lang]
+    llm = LLM(os.environ.get("MODEL", "Qwen/Qwen3.8-27B-FP8"), max_model_len=4096, gpu_memory_utilization=0.92, max_num_seqs=512,
+              limit_mm_per_prompt={"image": 0, "video": 0})
+    outs = llm.chat([[{"role": "user", "content": prompt(t, x)}] for t, x in zip(df.text, xl)],
+                    SamplingParams(temperature=0.7, top_p=0.95, max_tokens=300, seed=5 + sh), chat_template_kwargs={"enable_thinking": False})
+    os.makedirs(os.environ["OUT"], exist_ok=True); n = 0
+    with open(f"{os.environ['OUT']}/q_{sh}.jsonl", "w") as f:
+        for pid, lang, o in zip(df.pid, df.lang, outs):
+            j = parse_json(o.outputs[0].text)
+            if not isinstance(j, dict):
+                continue
+            for qt in ["question", "keywords", "english"]:
+                q = j.get(qt)
+                if isinstance(q, str) and 3 <= len(q.strip()) <= 300:
+                    f.write(json.dumps(dict(pid=int(pid), query=q.strip(), qtype=qt, qlang="eng_Latn" if qt == "english" else lang), ensure_ascii=False) + "\n"); n += 1
+    print("queries", n, "for", len(df), "passages", flush=True)
+    hard_exit(0)

config.json

@@ -0,0 +1,78 @@
+{
+  "architectures": [
+    "ModernBertModel"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 2,
+  "classifier_activation": "gelu",
+  "classifier_bias": false,
+  "classifier_dropout": 0.0,
+  "classifier_pooling": "mean",
+  "cls_token_id": 1,
+  "decoder_bias": true,
+  "deterministic_flash_attn": false,
+  "dtype": "float32",
+  "embedding_dropout": 0.0,
+  "eos_token_id": 1,
+  "global_attn_every_n_layers": 3,
+  "gradient_checkpointing": false,
+  "hidden_activation": "gelu",
+  "hidden_size": 768,
+  "initializer_cutoff_factor": 2.0,
+  "initializer_range": 0.02,
+  "intermediate_size": 1152,
+  "layer_norm_eps": 1e-05,
+  "layer_types": [
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "sliding_attention",
+    "full_attention"
+  ],
+  "local_attention": 128,
+  "mask_token_id": 4,
+  "max_position_embeddings": 8192,
+  "mlp_bias": false,
+  "mlp_dropout": 0.0,
+  "model_type": "modernbert",
+  "norm_bias": false,
+  "norm_eps": 1e-05,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 22,
+  "pad_token_id": 0,
+  "position_embedding_type": "sans_pos",
+  "rope_parameters": {
+    "full_attention": {
+      "rope_theta": 160000,
+      "rope_type": "default"
+    },
+    "sliding_attention": {
+      "rope_theta": 160000,
+      "rope_type": "default"
+    }
+  },
+  "sep_token_id": 1,
+  "sparse_pred_ignore_index": -100,
+  "sparse_prediction": false,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.17.0",
+  "vocab_size": 256000
+}

config_sentence_transformers.json

@@ -0,0 +1 @@
+{"similarity_fn_name": "cosine", "prompts": {}, "default_prompt_name": null}

eval/baselines.json

@@ -0,0 +1,857 @@
+{
+ "cls:BAAI/bge-m3": {
+  "askubuntu": {
+   "n": 361,
+   "ndcg10": 0.6656703810422914
+  },
+  "esci_es": {
+   "n": 150,
+   "ndcg10": 0.8353929290687163
+  },
+  "esci_jp": {
+   "n": 150,
+   "ndcg10": 0.8446809252178239
+  },
+  "esci_us": {
+   "n": 150,
+   "ndcg10": 0.8916635813106158
+  },
+  "miracl_ar": {
+   "n": 60,
+   "ndcg10": 0.8337544453555171
+  },
+  "miracl_bn": {
+   "n": 60,
+   "ndcg10": 0.8520468326002711
+  },
+  "miracl_de": {
+   "n": 60,
+   "ndcg10": 0.7296717789784314
+  },
+  "miracl_en": {
+   "n": 60,
+   "ndcg10": 0.6891871275339615
+  },
+  "miracl_es": {
+   "n": 60,
+   "ndcg10": 0.7581471021545895
+  },
+  "miracl_fa": {
+   "n": 60,
+   "ndcg10": 0.7846033120178257
+  },
+  "miracl_fi": {
+   "n": 60,
+   "ndcg10": 0.8202730056912013
+  },
+  "miracl_fr": {
+   "n": 60,
+   "ndcg10": 0.7004643716210734
+  },
+  "miracl_hi": {
+   "n": 60,
+   "ndcg10": 0.7604715990318043
+  },
+  "miracl_id": {
+   "n": 60,
+   "ndcg10": 0.6691216387932252
+  },
+  "miracl_ja": {
+   "n": 60,
+   "ndcg10": 0.7880794629562834
+  },
+  "miracl_ko": {
+   "n": 60,
+   "ndcg10": 0.7947333617834975
+  },
+  "miracl_ru": {
+   "n": 60,
+   "ndcg10": 0.8290247596475008
+  },
+  "miracl_sw": {
+   "n": 60,
+   "ndcg10": 0.8654063286257364
+  },
+  "miracl_te": {
+   "n": 60,
+   "ndcg10": 0.9329244281890277
+  },
+  "miracl_th": {
+   "n": 60,
+   "ndcg10": 0.8701537454495231
+  },
+  "miracl_yo": {
+   "n": 60,
+   "ndcg10": 0.870938042119561
+  },
+  "miracl_zh": {
+   "n": 60,
+   "ndcg10": 0.7891453552250133
+  },
+  "mmarco": {
+   "n": 150,
+   "ndcg10": 0.7100696637459445
+  },
+  "rubq": {
+   "n": 150,
+   "ndcg10": 0.8277617323379156
+  },
+  "t2": {
+   "n": 150,
+   "ndcg10": 0.7377569266122763
+  },
+  "wiki_bg": {
+   "n": 60,
+   "ndcg10": 0.929533513890146
+  },
+  "wiki_bn": {
+   "n": 60,
+   "ndcg10": 0.9108180336973224
+  },
+  "wiki_cs": {
+   "n": 60,
+   "ndcg10": 0.9553910969797961
+  },
+  "wiki_da": {
+   "n": 60,
+   "ndcg10": 0.9281230886209774
+  },
+  "wiki_de": {
+   "n": 60,
+   "ndcg10": 0.9520535437214951
+  },
+  "wiki_en": {
+   "n": 60,
+   "ndcg10": 0.9732131543452396
+  },
+  "wiki_fa": {
+   "n": 60,
+   "ndcg10": 0.9149356156733105
+  },
+  "wiki_fi": {
+   "n": 60,
+   "ndcg10": 0.9264564969075482
+  },
+  "wiki_hi": {
+   "n": 60,
+   "ndcg10": 0.8651733338339528
+  },
+  "wiki_it": {
+   "n": 60,
+   "ndcg10": 0.9356964284756129
+  },
+  "wiki_nl": {
+   "n": 60,
+   "ndcg10": 0.9587644899758274
+  },
+  "wiki_no": {
+   "n": 60,
+   "ndcg10": 0.8693061025833905
+  },
+  "wiki_pt": {
+   "n": 60,
+   "ndcg10": 0.9300560180372457
+  },
+  "wiki_ro": {
+   "n": 60,
+   "ndcg10": 0.947093603384193
+  },
+  "wiki_sr": {
+   "n": 60,
+   "ndcg10": 0.9265851157455774
+  },
+  "wiki_sv": {
+   "n": 60,
+   "ndcg10": 0.9141411337372607
+  },
+  "_miracl": 0.7965637054318913,
+  "_wiki": 0.927333798100556,
+  "_other": 0.7875708770479406,
+  "_all": 0.8371561268601293,
+  "_docs_per_s": 755.3071503037847
+ },
+ "e5:intfloat/multilingual-e5-small": {
+  "askubuntu": {
+   "n": 361,
+   "ndcg10": 0.6145866408660092
+  },
+  "esci_es": {
+   "n": 150,
+   "ndcg10": 0.8345672085259446
+  },
+  "esci_jp": {
+   "n": 150,
+   "ndcg10": 0.8386913872006542
+  },
+  "esci_us": {
+   "n": 150,
+   "ndcg10": 0.8975901738338674
+  },
+  "miracl_ar": {
+   "n": 60,
+   "ndcg10": 0.8094871343793595
+  },
+  "miracl_bn": {
+   "n": 60,
+   "ndcg10": 0.7114476822775221
+  },
+  "miracl_de": {
+   "n": 60,
+   "ndcg10": 0.6345644496075646
+  },
+  "miracl_en": {
+   "n": 60,
+   "ndcg10": 0.6191988201431645
+  },
+  "miracl_es": {
+   "n": 60,
+   "ndcg10": 0.7190329885079266
+  },
+  "miracl_fa": {
+   "n": 60,
+   "ndcg10": 0.6469699209144593
+  },
+  "miracl_fi": {
+   "n": 60,
+   "ndcg10": 0.7849846996428437
+  },
+  "miracl_fr": {
+   "n": 60,
+   "ndcg10": 0.5726760391631708
+  },
+  "miracl_hi": {
+   "n": 60,
+   "ndcg10": 0.6969545158673398
+  },
+  "miracl_id": {
+   "n": 60,
+   "ndcg10": 0.6005428547088218
+  },
+  "miracl_ja": {
+   "n": 60,
+   "ndcg10": 0.739137618018943
+  },
+  "miracl_ko": {
+   "n": 60,
+   "ndcg10": 0.7541787050586695
+  },
+  "miracl_ru": {
+   "n": 60,
+   "ndcg10": 0.745509866527369
+  },
+  "miracl_sw": {
+   "n": 60,
+   "ndcg10": 0.828129561398711
+  },
+  "miracl_te": {
+   "n": 60,
+   "ndcg10": 0.8646378971952645
+  },
+  "miracl_th": {
+   "n": 60,
+   "ndcg10": 0.7924128155944588
+  },
+  "miracl_yo": {
+   "n": 60,
+   "ndcg10": 0.8057363795220341
+  },
+  "miracl_zh": {
+   "n": 60,
+   "ndcg10": 0.6595681407849309
+  },
+  "mmarco": {
+   "n": 150,
+   "ndcg10": 0.7132981868433429
+  },
+  "rubq": {
+   "n": 150,
+   "ndcg10": 0.803404138209388
+  },
+  "t2": {
+   "n": 150,
+   "ndcg10": 0.7330930094196078
+  },
+  "wiki_bg": {
+   "n": 60,
+   "ndcg10": 0.8929525432074239
+  },
+  "wiki_bn": {
+   "n": 60,
+   "ndcg10": 0.9245838857688082
+  },
+  "wiki_cs": {
+   "n": 60,
+   "ndcg10": 0.9163377495151477
+  },
+  "wiki_da": {
+   "n": 60,
+   "ndcg10": 0.9330340941285132
+  },
+  "wiki_de": {
+   "n": 60,
+   "ndcg10": 0.9352582582167991
+  },
+  "wiki_en": {
+   "n": 60,
+   "ndcg10": 0.9720577636464628
+  },
+  "wiki_fa": {
+   "n": 60,
+   "ndcg10": 0.8823075178753691
+  },
+  "wiki_fi": {
+   "n": 60,
+   "ndcg10": 0.921830203917103
+  },
+  "wiki_hi": {
+   "n": 60,
+   "ndcg10": 0.8707020746341965
+  },
+  "wiki_it": {
+   "n": 60,
+   "ndcg10": 0.9267184647789447
+  },
+  "wiki_nl": {
+   "n": 60,
+   "ndcg10": 0.9266840566369079
+  },
+  "wiki_no": {
+   "n": 60,
+   "ndcg10": 0.8696024785968237
+  },
+  "wiki_pt": {
+   "n": 60,
+   "ndcg10": 0.9194119033063589
+  },
+  "wiki_ro": {
+   "n": 60,
+   "ndcg10": 0.9182876221499743
+  },
+  "wiki_sr": {
+   "n": 60,
+   "ndcg10": 0.9048929910571798
+  },
+  "wiki_sv": {
+   "n": 60,
+   "ndcg10": 0.9179815231135162
+  },
+  "_miracl": 0.7213983382951419,
+  "_wiki": 0.9145401956593455,
+  "_other": 0.7764615349855449,
+  "_all": 0.804133356313344,
+  "_docs_per_s": 1510.6246846209397
+ },
+ "e5:intfloat/multilingual-e5-base": {
+  "askubuntu": {
+   "n": 361,
+   "ndcg10": 0.6410350401013797
+  },
+  "esci_es": {
+   "n": 150,
+   "ndcg10": 0.8264960201468888
+  },
+  "esci_jp": {
+   "n": 150,
+   "ndcg10": 0.8527631585196286
+  },
+  "esci_us": {
+   "n": 150,
+   "ndcg10": 0.8979362353736146
+  },
+  "miracl_ar": {
+   "n": 60,
+   "ndcg10": 0.7741335914439994
+  },
+  "miracl_bn": {
+   "n": 60,
+   "ndcg10": 0.7470673775058853
+  },
+  "miracl_de": {
+   "n": 60,
+   "ndcg10": 0.6707476561103582
+  },
+  "miracl_en": {
+   "n": 60,
+   "ndcg10": 0.6434888531498094
+  },
+  "miracl_es": {
+   "n": 60,
+   "ndcg10": 0.7083119685592874
+  },
+  "miracl_fa": {
+   "n": 60,
+   "ndcg10": 0.7286007749434021
+  },
+  "miracl_fi": {
+   "n": 60,
+   "ndcg10": 0.769050059159852
+  },
+  "miracl_fr": {
+   "n": 60,
+   "ndcg10": 0.6438219522680882
+  },
+  "miracl_hi": {
+   "n": 60,
+   "ndcg10": 0.6879644832386964
+  },
+  "miracl_id": {
+   "n": 60,
+   "ndcg10": 0.5880643589648785
+  },
+  "miracl_ja": {
+   "n": 60,
+   "ndcg10": 0.7177894738320753
+  },
+  "miracl_ko": {
+   "n": 60,
+   "ndcg10": 0.733115368969295
+  },
+  "miracl_ru": {
+   "n": 60,
+   "ndcg10": 0.7557340800995906
+  },
+  "miracl_sw": {
+   "n": 60,
+   "ndcg10": 0.8440969533590856
+  },
+  "miracl_te": {
+   "n": 60,
+   "ndcg10": 0.8213691500565956
+  },
+  "miracl_th": {
+   "n": 60,
+   "ndcg10": 0.7775501105526337
+  },
+  "miracl_yo": {
+   "n": 60,
+   "ndcg10": 0.8602216409451844
+  },
+  "miracl_zh": {
+   "n": 60,
+   "ndcg10": 0.7718196626379744
+  },
+  "mmarco": {
+   "n": 150,
+   "ndcg10": 0.6860563287688956
+  },
+  "rubq": {
+   "n": 150,
+   "ndcg10": 0.8009261488182178
+  },
+  "t2": {
+   "n": 150,
+   "ndcg10": 0.7196655904910614
+  },
+  "wiki_bg": {
+   "n": 60,
+   "ndcg10": 0.888493096791215
+  },
+  "wiki_bn": {
+   "n": 60,
+   "ndcg10": 0.9053882337884093
+  },
+  "wiki_cs": {
+   "n": 60,
+   "ndcg10": 0.9353868770548284
+  },
+  "wiki_da": {
+   "n": 60,
+   "ndcg10": 0.9145106208218478
+  },
+  "wiki_de": {
+   "n": 60,
+   "ndcg10": 0.9382904104792485
+  },
+  "wiki_en": {
+   "n": 60,
+   "ndcg10": 0.9670619835714305
+  },
+  "wiki_fa": {
+   "n": 60,
+   "ndcg10": 0.8992370630146003
+  },
+  "wiki_fi": {
+   "n": 60,
+   "ndcg10": 0.9178611957028182
+  },
+  "wiki_hi": {
+   "n": 60,
+   "ndcg10": 0.8466475183304261
+  },
+  "wiki_it": {
+   "n": 60,
+   "ndcg10": 0.9428743711101694
+  },
+  "wiki_nl": {
+   "n": 60,
+   "ndcg10": 0.9461643408190826
+  },
+  "wiki_no": {
+   "n": 60,
+   "ndcg10": 0.8822254458686295
+  },
+  "wiki_pt": {
+   "n": 60,
+   "ndcg10": 0.93223818059677
+  },
+  "wiki_ro": {
+   "n": 60,
+   "ndcg10": 0.9240123664766273
+  },
+  "wiki_sr": {
+   "n": 60,
+   "ndcg10": 0.9236918945371214
+  },
+  "wiki_sv": {
+   "n": 60,
+   "ndcg10": 0.9298929910571798
+  },
+  "_miracl": 0.7357193064331494,
+  "_wiki": 0.9183735368762752,
+  "_other": 0.7749826460313837,
+  "_all": 0.8096918297802694,
+  "_docs_per_s": 1294.14767662913
+ },
+ "mean:sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2": {
+  "askubuntu": {
+   "n": 361,
+   "ndcg10": 0.6593311764088963
+  },
+  "esci_es": {
+   "n": 150,
+   "ndcg10": 0.8030820996228589
+  },
+  "esci_jp": {
+   "n": 150,
+   "ndcg10": 0.8292774648180925
+  },
+  "esci_us": {
+   "n": 150,
+   "ndcg10": 0.8558446274620043
+  },
+  "miracl_ar": {
+   "n": 60,
+   "ndcg10": 0.33995870060071187
+  },
+  "miracl_bn": {
+   "n": 60,
+   "ndcg10": 0.10062772888052027
+  },
+  "miracl_de": {
+   "n": 60,
+   "ndcg10": 0.49745229598874496
+  },
+  "miracl_en": {
+   "n": 60,
+   "ndcg10": 0.49430369069596836
+  },
+  "miracl_es": {
+   "n": 60,
+   "ndcg10": 0.49376054190150254
+  },
+  "miracl_fa": {
+   "n": 60,
+   "ndcg10": 0.33784773084632214
+  },
+  "miracl_fi": {
+   "n": 60,
+   "ndcg10": 0.5391379742395018
+  },
+  "miracl_fr": {
+   "n": 60,
+   "ndcg10": 0.4029746959581783
+  },
+  "miracl_hi": {
+   "n": 60,
+   "ndcg10": 0.3933734511901455
+  },
+  "miracl_id": {
+   "n": 60,
+   "ndcg10": 0.4195277671370329
+  },
+  "miracl_ja": {
+   "n": 60,
+   "ndcg10": 0.4299776871592214
+  },
+  "miracl_ko": {
+   "n": 60,
+   "ndcg10": 0.4232108800794022
+  },
+  "miracl_ru": {
+   "n": 60,
+   "ndcg10": 0.38301162077718887
+  },
+  "miracl_sw": {
+   "n": 60,
+   "ndcg10": 0.27333741060276756
+  },
+  "miracl_te": {
+   "n": 60,
+   "ndcg10": 0.10235710228305292
+  },
+  "miracl_th": {
+   "n": 60,
+   "ndcg10": 0.3751229772636128
+  },
+  "miracl_yo": {
+   "n": 60,
+   "ndcg10": 0.41564536084544285
+  },
+  "miracl_zh": {
+   "n": 60,
+   "ndcg10": 0.5455112257398346
+  },
+  "mmarco": {
+   "n": 150,
+   "ndcg10": 0.4529963705977627
+  },
+  "rubq": {
+   "n": 150,
+   "ndcg10": 0.5911557734552291
+  },
+  "t2": {
+   "n": 150,
+   "ndcg10": 0.6175956685016418
+  },
+  "wiki_bg": {
+   "n": 60,
+   "ndcg10": 0.8431693155279092
+  },
+  "wiki_bn": {
+   "n": 60,
+   "ndcg10": 0.51027776651553
+  },
+  "wiki_cs": {
+   "n": 60,
+   "ndcg10": 0.8739110090544683
+  },
+  "wiki_da": {
+   "n": 60,
+   "ndcg10": 0.8220310914235321
+  },
+  "wiki_de": {
+   "n": 60,
+   "ndcg10": 0.8539397656758437
+  },
+  "wiki_en": {
+   "n": 60,
+   "ndcg10": 0.9295452577018037
+  },
+  "wiki_fa": {
+   "n": 60,
+   "ndcg10": 0.8466261318110585
+  },
+  "wiki_fi": {
+   "n": 60,
+   "ndcg10": 0.8430048569521499
+  },
+  "wiki_hi": {
+   "n": 60,
+   "ndcg10": 0.7953517649819623
+  },
+  "wiki_it": {
+   "n": 60,
+   "ndcg10": 0.8847437240888024
+  },
+  "wiki_nl": {
+   "n": 60,
+   "ndcg10": 0.8861129781150623
+  },
+  "wiki_no": {
+   "n": 60,
+   "ndcg10": 0.8330340596148136
+  },
+  "wiki_pt": {
+   "n": 60,
+   "ndcg10": 0.8745424640104518
+  },
+  "wiki_ro": {
+   "n": 60,
+   "ndcg10": 0.8706507172922537
+  },
+  "wiki_sr": {
+   "n": 60,
+   "ndcg10": 0.8418999129941477
+  },
+  "wiki_sv": {
+   "n": 60,
+   "ndcg10": 0.8759933201214941
+  },
+  "_miracl": 0.3870632690105084,
+  "_wiki": 0.8365521334925803,
+  "_other": 0.687040454409498,
+  "_all": 0.636885285637529,
+  "_docs_per_s": 1368.597113329957
+ },
+ "mean:sentence-transformers/all-MiniLM-L6-v2": {
+  "askubuntu": {
+   "n": 361,
+   "ndcg10": 0.6950870859095775
+  },
+  "esci_es": {
+   "n": 150,
+   "ndcg10": 0.805459684054124
+  },
+  "esci_jp": {
+   "n": 150,
+   "ndcg10": 0.8190418405937759
+  },
+  "esci_us": {
+   "n": 150,
+   "ndcg10": 0.89287039530015
+  },
+  "miracl_ar": {
+   "n": 60,
+   "ndcg10": 0.038931957155848455
+  },
+  "miracl_bn": {
+   "n": 60,
+   "ndcg10": 0.030062987091536823
+  },
+  "miracl_de": {
+   "n": 60,
+   "ndcg10": 0.19863579759718686
+  },
+  "miracl_en": {
+   "n": 60,
+   "ndcg10": 0.44637962019197014
+  },
+  "miracl_es": {
+   "n": 60,
+   "ndcg10": 0.16533372301666774
+  },
+  "miracl_fa": {
+   "n": 60,
+   "ndcg10": 0.08099172914531616
+  },
+  "miracl_fi": {
+   "n": 60,
+   "ndcg10": 0.318827777529499
+  },
+  "miracl_fr": {
+   "n": 60,
+   "ndcg10": 0.19648428931651676
+  },
+  "miracl_hi": {
+   "n": 60,
+   "ndcg10": 0.05974075434463789
+  },
+  "miracl_id": {
+   "n": 60,
+   "ndcg10": 0.19480044817902895
+  },
+  "miracl_ja": {
+   "n": 60,
+   "ndcg10": 0.0664399201863372
+  },
+  "miracl_ko": {
+   "n": 60,
+   "ndcg10": 0.07768323456188891
+  },
+  "miracl_ru": {
+   "n": 60,
+   "ndcg10": 0.0741022641360989
+  },
+  "miracl_sw": {
+   "n": 60,
+   "ndcg10": 0.4118178738845216
+  },
+  "miracl_te": {
+   "n": 60,
+   "ndcg10": 0.028812270540518346
+  },
+  "miracl_th": {
+   "n": 60,
+   "ndcg10": 0.037664201840444354
+  },
+  "miracl_yo": {
+   "n": 60,
+   "ndcg10": 0.5977167422457427
+  },
+  "miracl_zh": {
+   "n": 60,
+   "ndcg10": 0.07071352407283164
+  },
+  "mmarco": {
+   "n": 150,
+   "ndcg10": 0.26867494304404316
+  },
+  "rubq": {
+   "n": 150,
+   "ndcg10": 0.3064855273291065
+  },
+  "t2": {
+   "n": 150,
+   "ndcg10": 0.5577994137229116
+  },
+  "wiki_bg": {
+   "n": 60,
+   "ndcg10": 0.6167824030088364
+  },
+  "wiki_bn": {
+   "n": 60,
+   "ndcg10": 0.4546528384572809
+  },
+  "wiki_cs": {
+   "n": 60,
+   "ndcg10": 0.7362788337728153
+  },
+  "wiki_da": {
+   "n": 60,
+   "ndcg10": 0.7773546957318817
+  },
+  "wiki_de": {
+   "n": 60,
+   "ndcg10": 0.7759664941138815
+  },
+  "wiki_en": {
+   "n": 60,
+   "ndcg10": 0.950431156642494
+  },
+  "wiki_fa": {
+   "n": 60,
+   "ndcg10": 0.5099313956928329
+  },
+  "wiki_fi": {
+   "n": 60,
+   "ndcg10": 0.801663620538976
+  },
+  "wiki_hi": {
+   "n": 60,
+   "ndcg10": 0.5357898617158996
+  },
+  "wiki_it": {
+   "n": 60,
+   "ndcg10": 0.8319465347786749
+  },
+  "wiki_nl": {
+   "n": 60,
+   "ndcg10": 0.8007722530139283
+  },
+  "wiki_no": {
+   "n": 60,
+   "ndcg10": 0.7436514286534031
+  },
+  "wiki_pt": {
+   "n": 60,
+   "ndcg10": 0.8185917340825137
+  },
+  "wiki_ro": {
+   "n": 60,
+   "ndcg10": 0.8534025214158526
+  },
+  "wiki_sr": {
+   "n": 60,
+   "ndcg10": 0.5560142087611005
+  },
+  "wiki_sv": {
+   "n": 60,
+   "ndcg10": 0.7582896375765296
+  },
+  "_miracl": 0.17195217305758848,
+  "_wiki": 0.7200949761223062,
+  "_other": 0.6207741271362411,
+  "_all": 0.5042737587720453,
+  "_docs_per_s": 1423.8455996927694
+ }
+}

eval/this_model.json

@@ -0,0 +1,344 @@
+{
+ "ours:/fast/awuhrmann/agent-outputs/run-20260923-130458/9ab0016a-b6ee-4764-994d-5154baf33dc7/model": {
+  "askubuntu": {
+   "n": 361,
+   "ndcg10": 0.6454024803892467
+  },
+  "esci_es": {
+   "n": 150,
+   "ndcg10": 0.8421130165954218
+  },
+  "esci_jp": {
+   "n": 150,
+   "ndcg10": 0.8485930209752721
+  },
+  "esci_us": {
+   "n": 150,
+   "ndcg10": 0.880127369013804
+  },
+  "miracl_ar": {
+   "n": 60,
+   "ndcg10": 0.8217118082576983
+  },
+  "miracl_bn": {
+   "n": 60,
+   "ndcg10": 0.799904189410707
+  },
+  "miracl_de": {
+   "n": 60,
+   "ndcg10": 0.6870982746080678
+  },
+  "miracl_en": {
+   "n": 60,
+   "ndcg10": 0.6669980874794317
+  },
+  "miracl_es": {
+   "n": 60,
+   "ndcg10": 0.7529764511365112
+  },
+  "miracl_fa": {
+   "n": 60,
+   "ndcg10": 0.7224917329012956
+  },
+  "miracl_fi": {
+   "n": 60,
+   "ndcg10": 0.8109735726187206
+  },
+  "miracl_fr": {
+   "n": 60,
+   "ndcg10": 0.7028965920193273
+  },
+  "miracl_hi": {
+   "n": 60,
+   "ndcg10": 0.6579653261024776
+  },
+  "miracl_id": {
+   "n": 60,
+   "ndcg10": 0.6438096608032022
+  },
+  "miracl_ja": {
+   "n": 60,
+   "ndcg10": 0.7225855446687406
+  },
+  "miracl_ko": {
+   "n": 60,
+   "ndcg10": 0.7598925416390455
+  },
+  "miracl_ru": {
+   "n": 60,
+   "ndcg10": 0.7720009403332085
+  },
+  "miracl_sw": {
+   "n": 60,
+   "ndcg10": 0.7914829344017279
+  },
+  "miracl_te": {
+   "n": 60,
+   "ndcg10": 0.8415466359969032
+  },
+  "miracl_th": {
+   "n": 60,
+   "ndcg10": 0.7793873296707757
+  },
+  "miracl_yo": {
+   "n": 60,
+   "ndcg10": 0.8945753374362755
+  },
+  "miracl_zh": {
+   "n": 60,
+   "ndcg10": 0.7285885887686003
+  },
+  "mmarco": {
+   "n": 150,
+   "ndcg10": 0.6673208259725423
+  },
+  "rubq": {
+   "n": 150,
+   "ndcg10": 0.8145785161115712
+  },
+  "t2": {
+   "n": 150,
+   "ndcg10": 0.7389363486924803
+  },
+  "wiki_bg": {
+   "n": 60,
+   "ndcg10": 0.9488921928506111
+  },
+  "wiki_bn": {
+   "n": 60,
+   "ndcg10": 0.9464263086904791
+  },
+  "wiki_cs": {
+   "n": 60,
+   "ndcg10": 0.9553910969797961
+  },
+  "wiki_da": {
+   "n": 60,
+   "ndcg10": 0.9725185982356556
+  },
+  "wiki_de": {
+   "n": 60,
+   "ndcg10": 0.9401761970253395
+  },
+  "wiki_en": {
+   "n": 60,
+   "ndcg10": 0.9430887554321781
+  },
+  "wiki_fa": {
+   "n": 60,
+   "ndcg10": 0.8937818799460686
+  },
+  "wiki_fi": {
+   "n": 60,
+   "ndcg10": 0.9587286502380972
+  },
+  "wiki_hi": {
+   "n": 60,
+   "ndcg10": 0.8568689804067117
+  },
+  "wiki_it": {
+   "n": 60,
+   "ndcg10": 0.9396795226984166
+  },
+  "wiki_nl": {
+   "n": 60,
+   "ndcg10": 0.9516840566369079
+  },
+  "wiki_no": {
+   "n": 60,
+   "ndcg10": 0.8828827306447898
+  },
+  "wiki_pt": {
+   "n": 60,
+   "ndcg10": 0.9383893513705791
+  },
+  "wiki_ro": {
+   "n": 60,
+   "ndcg10": 0.9399618127033308
+  },
+  "wiki_sr": {
+   "n": 60,
+   "ndcg10": 0.9659065928726538
+  },
+  "wiki_sv": {
+   "n": 60,
+   "ndcg10": 0.9096481897801829
+  },
+  "_miracl": 0.7531603082362621,
+  "_wiki": 0.9340015572819873,
+  "_other": 0.7767245111071912,
+  "_all": 0.8212954588751469,
+  "_docs_per_s": 871.5669460239191
+ },
+ "asym:/fast/awuhrmann/agent-outputs/run-20260923-130458/9ab0016a-b6ee-4764-994d-5154baf33dc7/model": {
+  "askubuntu": {
+   "n": 361,
+   "ndcg10": 0.6440774120514842
+  },
+  "esci_es": {
+   "n": 150,
+   "ndcg10": 0.8385399733212493
+  },
+  "esci_jp": {
+   "n": 150,
+   "ndcg10": 0.8562850126133227
+  },
+  "esci_us": {
+   "n": 150,
+   "ndcg10": 0.8862480984456198
+  },
+  "miracl_ar": {
+   "n": 60,
+   "ndcg10": 0.7954835090989633
+  },
+  "miracl_bn": {
+   "n": 60,
+   "ndcg10": 0.7815916462095401
+  },
+  "miracl_de": {
+   "n": 60,
+   "ndcg10": 0.7076742754712886
+  },
+  "miracl_en": {
+   "n": 60,
+   "ndcg10": 0.6700312003323375
+  },
+  "miracl_es": {
+   "n": 60,
+   "ndcg10": 0.7095367439753345
+  },
+  "miracl_fa": {
+   "n": 60,
+   "ndcg10": 0.6867360199962332
+  },
+  "miracl_fi": {
+   "n": 60,
+   "ndcg10": 0.8020650186446121
+  },
+  "miracl_fr": {
+   "n": 60,
+   "ndcg10": 0.6765686304868476
+  },
+  "miracl_hi": {
+   "n": 60,
+   "ndcg10": 0.6453188140969595
+  },
+  "miracl_id": {
+   "n": 60,
+   "ndcg10": 0.6360307891031997
+  },
+  "miracl_ja": {
+   "n": 60,
+   "ndcg10": 0.7344066663358962
+  },
+  "miracl_ko": {
+   "n": 60,
+   "ndcg10": 0.7342229813172885
+  },
+  "miracl_ru": {
+   "n": 60,
+   "ndcg10": 0.8072169697526407
+  },
+  "miracl_sw": {
+   "n": 60,
+   "ndcg10": 0.8259876439612289
+  },
+  "miracl_te": {
+   "n": 60,
+   "ndcg10": 0.8189061813409654
+  },
+  "miracl_th": {
+   "n": 60,
+   "ndcg10": 0.7803309379115535
+  },
+  "miracl_yo": {
+   "n": 60,
+   "ndcg10": 0.8552227103388154
+  },
+  "miracl_zh": {
+   "n": 60,
+   "ndcg10": 0.6921012777903786
+  },
+  "mmarco": {
+   "n": 150,
+   "ndcg10": 0.6369943090342021
+  },
+  "rubq": {
+   "n": 150,
+   "ndcg10": 0.8374024209496113
+  },
+  "t2": {
+   "n": 150,
+   "ndcg10": 0.7203977984132124
+  },
+  "wiki_bg": {
+   "n": 60,
+   "ndcg10": 0.957835227410717
+  },
+  "wiki_bn": {
+   "n": 60,
+   "ndcg10": 0.9295452577018037
+  },
+  "wiki_cs": {
+   "n": 60,
+   "ndcg10": 0.9676934385274142
+  },
+  "wiki_da": {
+   "n": 60,
+   "ndcg10": 0.945602223141698
+  },
+  "wiki_de": {
+   "n": 60,
+   "ndcg10": 0.9527749145863909
+  },
+  "wiki_en": {
+   "n": 60,
+   "ndcg10": 0.9587286502380972
+  },
+  "wiki_fa": {
+   "n": 60,
+   "ndcg10": 0.8920367429441932
+  },
+  "wiki_fi": {
+   "n": 60,
+   "ndcg10": 0.9613278834315966
+  },
+  "wiki_hi": {
+   "n": 60,
+   "ndcg10": 0.8604873920742336
+  },
+  "wiki_it": {
+   "n": 60,
+   "ndcg10": 0.9391197472178933
+  },
+  "wiki_nl": {
+   "n": 60,
+   "ndcg10": 0.9428743711101694
+  },
+  "wiki_no": {
+   "n": 60,
+   "ndcg10": 0.8924789738900906
+  },
+  "wiki_pt": {
+   "n": 60,
+   "ndcg10": 0.9261228495606912
+  },
+  "wiki_ro": {
+   "n": 60,
+   "ndcg10": 0.933600031400068
+  },
+  "wiki_sr": {
+   "n": 60,
+   "ndcg10": 0.9615422677536052
+  },
+  "wiki_sv": {
+   "n": 60,
+   "ndcg10": 0.9181101419515455
+  },
+  "_miracl": 0.7421906675646712,
+  "_wiki": 0.933742507058763,
+  "_other": 0.7742778606898144,
+  "_all": 0.8167370117710829,
+  "_docs_per_s": 789.6014879056063
+ }
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4492406536ba29beaaa21b891a13442a4ca81d3c7c96338ab9fec6aa599b5aaa
+size 1227771776

modules.json

@@ -0,0 +1,26 @@
+[
+ {
+  "idx": 0,
+  "name": "0",
+  "path": "",
+  "type": "sentence_transformers.models.Transformer"
+ },
+ {
+  "idx": 1,
+  "name": "1",
+  "path": "1_Pooling",
+  "type": "sentence_transformers.models.Pooling"
+ },
+ {
+  "idx": 2,
+  "name": "2",
+  "path": "2_Dense",
+  "type": "sentence_transformers.models.Dense"
+ },
+ {
+  "idx": 3,
+  "name": "3",
+  "path": "3_Normalize",
+  "type": "sentence_transformers.models.Normalize"
+ }
+]

onnx/model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4faca1585b2bb2ceb5443aa13e4ed543f9148a7f81bd4808cd4bb813cd0d82f7
+size 1231445467

onnx/model_quantized.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7aace31160d136c30afca8e42de7dd398a8e62ab192063c735bb16886bb71b7d
+size 641822905

onnx/quantization_check.json

@@ -0,0 +1,11 @@
+{
+ "gather_only": {
+  "min_cos": 0.9999687671661377,
+  "mb": 641.822905
+ },
+ "fp32": {
+  "cos_to_torch": 0.9999997615814209,
+  "mb": 1231.445467
+ },
+ "n": 4
+}

sentence_bert_config.json

@@ -0,0 +1 @@
+{"max_seq_length": 512, "do_lower_case": false}

special_tokens_map.json

@@ -0,0 +1,34 @@
+{
+  "additional_special_tokens": [
+    "<start_of_turn>",
+    "<end_of_turn>"
+  ],
+  "bos_token": {
+    "content": "<bos>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<eos>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<pad>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a5069feab370ece9f452dcda41c0ba2c216eebf6cb49b06893698f53cdc2b0b
+size 34363441

tokenizer_config.json

@@ -0,0 +1,2017 @@
+{
+  "add_bos_token": true,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<pad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<eos>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "<bos>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "4": {
+      "content": "<mask>",
+      "lstrip": true,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "5": {
+      "content": "<2mass>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "6": {
+      "content": "[@BOS@]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "7": {
+      "content": "<unused0>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "8": {
+      "content": "<unused1>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "9": {
+      "content": "<unused2>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "10": {
+      "content": "<unused3>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "11": {
+      "content": "<unused4>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "12": {
+      "content": "<unused5>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "13": {
+      "content": "<unused6>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "14": {
+      "content": "<unused7>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "15": {
+      "content": "<unused8>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "16": {
+      "content": "<unused9>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "17": {
+      "content": "<unused10>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "18": {
+      "content": "<unused11>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "19": {
+      "content": "<unused12>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "20": {
+      "content": "<unused13>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "21": {
+      "content": "<unused14>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "22": {
+      "content": "<unused15>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "23": {
+      "content": "<unused16>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "24": {
+      "content": "<unused17>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "25": {
+      "content": "<unused18>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "26": {
+      "content": "<unused19>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "27": {
+      "content": "<unused20>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "28": {
+      "content": "<unused21>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "29": {
+      "content": "<unused22>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "30": {
+      "content": "<unused23>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "31": {
+      "content": "<unused24>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "32": {
+      "content": "<unused25>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "33": {
+      "content": "<unused26>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "34": {
+      "content": "<unused27>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "35": {
+      "content": "<unused28>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "36": {
+      "content": "<unused29>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "37": {
+      "content": "<unused30>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "38": {
+      "content": "<unused31>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "39": {
+      "content": "<unused32>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "40": {
+      "content": "<unused33>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "41": {
+      "content": "<unused34>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "42": {
+      "content": "<unused35>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "43": {
+      "content": "<unused36>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "44": {
+      "content": "<unused37>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "45": {
+      "content": "<unused38>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "46": {
+      "content": "<unused39>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "47": {
+      "content": "<unused40>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "48": {
+      "content": "<unused41>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "49": {
+      "content": "<unused42>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50": {
+      "content": "<unused43>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "51": {
+      "content": "<unused44>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "52": {
+      "content": "<unused45>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "53": {
+      "content": "<unused46>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "54": {
+      "content": "<unused47>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "55": {
+      "content": "<unused48>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "56": {
+      "content": "<unused49>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "57": {
+      "content": "<unused50>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "58": {
+      "content": "<unused51>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "59": {
+      "content": "<unused52>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "60": {
+      "content": "<unused53>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "61": {
+      "content": "<unused54>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "62": {
+      "content": "<unused55>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "63": {
+      "content": "<unused56>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "64": {
+      "content": "<unused57>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "65": {
+      "content": "<unused58>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "66": {
+      "content": "<unused59>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "67": {
+      "content": "<unused60>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "68": {
+      "content": "<unused61>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "69": {
+      "content": "<unused62>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "70": {
+      "content": "<unused63>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "71": {
+      "content": "<unused64>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "72": {
+      "content": "<unused65>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "73": {
+      "content": "<unused66>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "74": {
+      "content": "<unused67>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "75": {
+      "content": "<unused68>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "76": {
+      "content": "<unused69>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "77": {
+      "content": "<unused70>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "78": {
+      "content": "<unused71>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "79": {
+      "content": "<unused72>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "80": {
+      "content": "<unused73>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "81": {
+      "content": "<unused74>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "82": {
+      "content": "<unused75>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "83": {
+      "content": "<unused76>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "84": {
+      "content": "<unused77>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "85": {
+      "content": "<unused78>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "86": {
+      "content": "<unused79>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "87": {
+      "content": "<unused80>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "88": {
+      "content": "<unused81>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "89": {
+      "content": "<unused82>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "90": {
+      "content": "<unused83>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "91": {
+      "content": "<unused84>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "92": {
+      "content": "<unused85>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "93": {
+      "content": "<unused86>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "94": {
+      "content": "<unused87>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "95": {
+      "content": "<unused88>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "96": {
+      "content": "<unused89>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "97": {
+      "content": "<unused90>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "98": {
+      "content": "<unused91>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "99": {
+      "content": "<unused92>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "100": {
+      "content": "<unused93>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "101": {
+      "content": "<unused94>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "102": {
+      "content": "<unused95>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "103": {
+      "content": "<unused96>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "104": {
+      "content": "<unused97>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "105": {
+      "content": "<unused98>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "106": {
+      "content": "<start_of_turn>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "107": {
+      "content": "<end_of_turn>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "108": {
+      "content": "\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "109": {
+      "content": "\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "110": {
+      "content": "\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "111": {
+      "content": "\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "112": {
+      "content": "\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "113": {
+      "content": "\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "114": {
+      "content": "\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "115": {
+      "content": "\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "116": {
+      "content": "\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "117": {
+      "content": "\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "118": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "119": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "120": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "121": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "122": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "123": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "124": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "125": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "126": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "127": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "128": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "129": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "130": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "131": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "132": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "133": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "134": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "135": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "136": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "137": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "138": {
+      "content": "\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "139": {
+      "content": "▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "140": {
+      "content": "▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "141": {
+      "content": "▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "142": {
+      "content": "▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "143": {
+      "content": "▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "144": {
+      "content": "▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "145": {
+      "content": "▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "146": {
+      "content": "▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "147": {
+      "content": "▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "148": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "149": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "150": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "152": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "153": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "154": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "155": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "156": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "157": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "158": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "159": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "160": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "161": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "162": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "163": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "164": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "165": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "166": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "167": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "168": {
+      "content": "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "169": {
+      "content": "<table>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "170": {
+      "content": "<caption>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "171": {
+      "content": "<thead>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "172": {
+      "content": "<tbody>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "173": {
+      "content": "<tfoot>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "174": {
+      "content": "<tr>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "175": {
+      "content": "<th>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "176": {
+      "content": "<td>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "177": {
+      "content": "</table>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "178": {
+      "content": "</caption>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "179": {
+      "content": "</thead>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "180": {
+      "content": "</tbody>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "181": {
+      "content": "</tfoot>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "182": {
+      "content": "</tr>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "183": {
+      "content": "</th>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "184": {
+      "content": "</td>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "185": {
+      "content": "<h1>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "186": {
+      "content": "<h2>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "187": {
+      "content": "<h3>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "188": {
+      "content": "<h4>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "189": {
+      "content": "<h5>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "190": {
+      "content": "<h6>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "191": {
+      "content": "<blockquote>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "192": {
+      "content": "</h1>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "193": {
+      "content": "</h2>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "194": {
+      "content": "</h3>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "195": {
+      "content": "</h4>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "196": {
+      "content": "</h5>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "197": {
+      "content": "</h6>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "198": {
+      "content": "</blockquote>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "199": {
+      "content": "<strong>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "200": {
+      "content": "<em>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "201": {
+      "content": "<b>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "202": {
+      "content": "<i>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "203": {
+      "content": "<u>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "204": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "205": {
+      "content": "<sub>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "206": {
+      "content": "<sup>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "207": {
+      "content": "<code>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "208": {
+      "content": "</strong>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "209": {
+      "content": "</em>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "210": {
+      "content": "</b>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "211": {
+      "content": "</i>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "212": {
+      "content": "</u>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "213": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "214": {
+      "content": "</sub>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "215": {
+      "content": "</sup>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "216": {
+      "content": "</code>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255968": {
+      "content": "[toxicity=0]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255969": {
+      "content": "\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255970": {
+      "content": "\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255971": {
+      "content": "\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255972": {
+      "content": "\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255973": {
+      "content": "\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255974": {
+      "content": "\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255975": {
+      "content": "\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255976": {
+      "content": "\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255977": {
+      "content": "\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255978": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255979": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255980": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255981": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255982": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255983": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255984": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255985": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255986": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255987": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255988": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255989": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255990": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255991": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255992": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255993": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255994": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255995": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255996": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255997": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255998": {
+      "content": "\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "255999": {
+      "content": "<unused99>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<start_of_turn>",
+    "<end_of_turn>"
+  ],
+  "padding_side": "right",
+  "bos_token": "<bos>",
+  "cls_token": "<bos>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<eos>",
+  "sep_token": "<eos>",
+  "model_max_length": 8192,
+  "pad_token": "<pad>",
+  "mask_token": "<mask>",
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "PreTrainedTokenizerFast",
+  "unk_token": "<unk>",
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ]
+}

training/data_stats.json

@@ -0,0 +1,8 @@
+{
+ "distillation_texts": 4650511,
+ "fineweb2_snippets": 3397742,
+ "rerank_passages": 399754,
+ "generated_queries": 853015,
+ "languages": 92,
+ "teacher": "BAAI/bge-m3 (dense, CLS, normalised)"
+}

training/train_log.json

@@ -0,0 +1,43 @@
+{
+ "args": {
+  "base": "jhu-clsp/mmBERT-base",
+  "teacher": "see code/",
+  "out": "model",
+  "max_len": 256,
+  "bs": 256,
+  "lr": 5e-05,
+  "epochs": 1.0,
+  "warmup": 0.03,
+  "mse_w": 1.0,
+  "sim_w": 1.0,
+  "seed": 0,
+  "eval_every": 4000
+ },
+ "log": [
+  {
+   "step": 4000,
+   "cos": 0.7902698516845703,
+   "top1": 0.999267578125
+  },
+  {
+   "step": 8000,
+   "cos": 0.8306772708892822,
+   "top1": 1.0
+  },
+  {
+   "step": 12000,
+   "cos": 0.8455853462219238,
+   "top1": 1.0
+  },
+  {
+   "step": 16000,
+   "cos": 0.8512239456176758,
+   "top1": 1.0
+  },
+  {
+   "step": 18074,
+   "cos": 0.851382851600647,
+   "top1": 1.0
+  }
+ ]
+}

training/val_metrics.json

@@ -0,0 +1,5 @@
+{
+ "step": 18074,
+ "cos": 0.851382851600647,
+ "top1": 1.0
+}