Add new SentenceTransformer model

.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": 1024,
+  "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
+}

config.json

@@ -0,0 +1,67 @@
+{
+  "_name_or_path": "jinaai/jina-embeddings-v3",
+  "architectures": [
+    "XLMRobertaLoRA"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "auto_map": {
+    "AutoConfig": "jinaai/xlm-roberta-flash-implementation--configuration_xlm_roberta.XLMRobertaFlashConfig",
+    "AutoModel": "jinaai/xlm-roberta-flash-implementation--modeling_lora.XLMRobertaLoRA",
+    "AutoModelForMaskedLM": "jinaai/xlm-roberta-flash-implementation--modeling_xlm_roberta.XLMRobertaForMaskedLM",
+    "AutoModelForPreTraining": "jinaai/xlm-roberta-flash-implementation--modeling_xlm_roberta.XLMRobertaForPreTraining"
+  },
+  "bos_token_id": 0,
+  "classifier_dropout": null,
+  "emb_pooler": null,
+  "eos_token_id": 2,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "load_trained_adapters": true,
+  "lora_adaptations": [
+    "retrieval.query",
+    "retrieval.passage",
+    "separation",
+    "classification",
+    "text-matching"
+  ],
+  "lora_alpha": 1,
+  "lora_dropout_p": 0.0,
+  "lora_main_params_trainable": false,
+  "lora_rank": 4,
+  "matryoshka_dimensions": [
+    32,
+    64,
+    128,
+    256,
+    512,
+    768,
+    1024
+  ],
+  "max_position_embeddings": 8194,
+  "model_type": "xlm-roberta",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "output_past": true,
+  "pad_token_id": 1,
+  "position_embedding_type": "rotary",
+  "rotary_emb_base": 20000.0,
+  "task_instructions": {
+    "classification": "",
+    "retrieval.passage": "Represent the document for retrieval: ",
+    "retrieval.query": "Represent the query for retrieving evidence documents: ",
+    "separation": "",
+    "text-matching": ""
+  },
+  "torch_dtype": "float32",
+  "transformers_version": "4.48.0",
+  "truncate_dim": null,
+  "type_vocab_size": 1,
+  "use_cache": true,
+  "use_flash_attn": true,
+  "use_reentrant": false,
+  "vocab_size": 250002
+}

config_sentence_transformers.json

@@ -0,0 +1,16 @@
+{
+  "__version__": {
+    "sentence_transformers": "3.4.1",
+    "transformers": "4.48.0",
+    "pytorch": "2.1.0"
+  },
+  "prompts": {
+    "retrieval.query": "Represent the query for retrieving evidence documents: ",
+    "retrieval.passage": "Represent the document for retrieval: ",
+    "separation": "",
+    "classification": "",
+    "text-matching": ""
+  },
+  "default_prompt_name": null,
+  "similarity_fn_name": "cosine"
+}

custom_st.py

@@ -0,0 +1,229 @@
+import json
+import logging
+import os
+from io import BytesIO
+from typing import Any, Dict, List, Optional, Tuple, Union
+
+import torch
+from torch import nn
+from transformers import AutoConfig, AutoModel, AutoTokenizer
+
+logger = logging.getLogger(__name__)
+
+
+class Transformer(nn.Module):
+    """Huggingface AutoModel to generate token embeddings.
+    Loads the correct class, e.g. BERT / RoBERTa etc.
+
+    Args:
+        model_name_or_path: Huggingface models name
+            (https://huggingface.co/models)
+        max_seq_length: Truncate any inputs longer than max_seq_length
+        model_args: Keyword arguments passed to the Huggingface
+            Transformers model
+        tokenizer_args: Keyword arguments passed to the Huggingface
+            Transformers tokenizer
+        config_args: Keyword arguments passed to the Huggingface
+            Transformers config
+        cache_dir: Cache dir for Huggingface Transformers to store/load
+            models
+        do_lower_case: If true, lowercases the input (independent if the
+            model is cased or not)
+        tokenizer_name_or_path: Name or path of the tokenizer. When
+            None, then model_name_or_path is used
+    """
+
+    save_in_root: bool = True
+
+    def __init__(
+        self,
+        model_name_or_path: str,
+        max_seq_length: int = None,
+        model_args: Dict[str, Any] = None,
+        tokenizer_args: Dict[str, Any] = None,
+        config_args: Dict[str, Any] = None,
+        cache_dir: str = None,
+        do_lower_case: bool = False,
+        tokenizer_name_or_path: str = None,
+        **kwargs,
+    ) -> None:
+        super().__init__()
+        self.config_keys = ["max_seq_length", "do_lower_case"]
+        self.do_lower_case = do_lower_case
+        if model_args is None:
+            model_args = {}
+        if tokenizer_args is None:
+            tokenizer_args = {}
+        if config_args is None:
+            config_args = {}
+
+        if kwargs.get("backend", "torch") != "torch":
+            logger.warning(
+                f'"jinaai/jina-embeddings-v3" is currently not compatible with the {kwargs["backend"]} backend. '
+                'Continuing with the "torch" backend.'
+            )
+        
+        self.config = AutoConfig.from_pretrained(model_name_or_path, **config_args, cache_dir=cache_dir)
+
+        self._lora_adaptations = self.config.lora_adaptations
+        if (
+            not isinstance(self._lora_adaptations, list)
+            or len(self._lora_adaptations) < 1
+        ):
+            raise ValueError(
+                f"`lora_adaptations` must be a list and contain at least one element"
+            )
+        self._adaptation_map = {
+            name: idx for idx, name in enumerate(self._lora_adaptations)
+        }
+
+        self.default_task = model_args.pop('default_task', None)
+
+        self.auto_model = AutoModel.from_pretrained(model_name_or_path, config=self.config, cache_dir=cache_dir, **model_args)
+
+        if max_seq_length is not None and "model_max_length" not in tokenizer_args:
+            tokenizer_args["model_max_length"] = max_seq_length
+        self.tokenizer = AutoTokenizer.from_pretrained(
+            tokenizer_name_or_path if tokenizer_name_or_path is not None else model_name_or_path,
+            cache_dir=cache_dir,
+            **tokenizer_args,
+        )
+
+        # No max_seq_length set. Try to infer from model
+        if max_seq_length is None:
+            if (
+                hasattr(self.auto_model, "config")
+                and hasattr(self.auto_model.config, "max_position_embeddings")
+                and hasattr(self.tokenizer, "model_max_length")
+            ):
+                max_seq_length = min(self.auto_model.config.max_position_embeddings, self.tokenizer.model_max_length)
+
+        self.max_seq_length = max_seq_length
+
+        if tokenizer_name_or_path is not None:
+            self.auto_model.config.tokenizer_class = self.tokenizer.__class__.__name__
+
+
+    @property
+    def default_task(self):
+        return self._default_task
+
+    @default_task.setter
+    def default_task(self, task: Union[None, str]):
+        self._validate_task(task)
+        self._default_task = task
+        
+
+    def _validate_task(self, task: str):
+        if task and task not in self._lora_adaptations:
+            raise ValueError(
+                f"Unsupported task '{task}'. "
+                f"Supported tasks are: {', '.join(self.config.lora_adaptations)}. "
+                f"Alternatively, don't pass the `task` argument to disable LoRA."
+            )
+
+    def forward(
+        self, features: Dict[str, torch.Tensor], task: Optional[str] = None
+    ) -> Dict[str, torch.Tensor]:
+        """Returns token_embeddings, cls_token"""
+        self._validate_task(task)
+        task = task or self.default_task
+        adapter_mask = None
+        if task:
+            task_id = self._adaptation_map[task]
+            num_examples = features['input_ids'].size(0)
+            adapter_mask = torch.full(
+                (num_examples,), task_id, dtype=torch.int32, device=features['input_ids'].device
+            )
+
+        lora_arguments = (
+            {"adapter_mask": adapter_mask} if adapter_mask is not None else {}
+        )
+        features.pop('prompt_length', None)
+        output_states = self.auto_model.forward(**features, **lora_arguments, return_dict=False)
+        output_tokens = output_states[0]
+        features.update({"token_embeddings": output_tokens, "attention_mask": features["attention_mask"]})
+        return features
+
+    def get_word_embedding_dimension(self) -> int:
+        return self.auto_model.config.hidden_size
+
+    def tokenize(
+        self,
+        texts: Union[List[str], List[dict], List[Tuple[str, str]]],
+        padding: Union[str, bool] = True
+    ) -> Dict[str, torch.Tensor]:
+        """Tokenizes a text and maps tokens to token-ids"""
+        output = {}
+        if isinstance(texts[0], str):
+            to_tokenize = [texts]
+        elif isinstance(texts[0], dict):
+            to_tokenize = []
+            output["text_keys"] = []
+            for lookup in texts:
+                text_key, text = next(iter(lookup.items()))
+                to_tokenize.append(text)
+                output["text_keys"].append(text_key)
+            to_tokenize = [to_tokenize]
+        else:
+            batch1, batch2 = [], []
+            for text_tuple in texts:
+                batch1.append(text_tuple[0])
+                batch2.append(text_tuple[1])
+            to_tokenize = [batch1, batch2]
+
+        # strip
+        to_tokenize = [[str(s).strip() for s in col] for col in to_tokenize]
+
+        # Lowercase
+        if self.do_lower_case:
+            to_tokenize = [[s.lower() for s in col] for col in to_tokenize]
+
+        output.update(
+            self.tokenizer(
+                *to_tokenize,
+                padding=padding,
+                truncation="longest_first",
+                return_tensors="pt",
+                max_length=self.max_seq_length,
+            )
+        )
+        return output
+
+    def get_config_dict(self) -> Dict[str, Any]:
+         return {key: self.__dict__[key] for key in self.config_keys}
+
+    def save(self, output_path: str, safe_serialization: bool = True) -> None:
+        self.auto_model.save_pretrained(output_path, safe_serialization=safe_serialization)
+        self.tokenizer.save_pretrained(output_path)
+
+        with open(os.path.join(output_path, "sentence_bert_config.json"), "w") as fOut:
+            json.dump(self.get_config_dict(), fOut, indent=2)
+
+
+    @classmethod
+    def load(cls, input_path: str) -> "Transformer":
+        # Old classes used other config names than 'sentence_bert_config.json'
+        for config_name in [
+            "sentence_bert_config.json",
+            "sentence_roberta_config.json",
+            "sentence_distilbert_config.json",
+            "sentence_camembert_config.json",
+            "sentence_albert_config.json",
+            "sentence_xlm-roberta_config.json",
+            "sentence_xlnet_config.json",
+        ]:
+            sbert_config_path = os.path.join(input_path, config_name)
+            if os.path.exists(sbert_config_path):
+                break
+
+        with open(sbert_config_path) as fIn:
+            config = json.load(fIn)
+        # Don't allow configs to set trust_remote_code
+        if "model_args" in config and "trust_remote_code" in config["model_args"]:
+            config["model_args"].pop("trust_remote_code")
+        if "tokenizer_args" in config and "trust_remote_code" in config["tokenizer_args"]:
+            config["tokenizer_args"].pop("trust_remote_code")
+        if "config_args" in config and "trust_remote_code" in config["config_args"]:
+            config["config_args"].pop("trust_remote_code")
+        return cls(model_name_or_path=input_path, **config)

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6658a34d2f9999b96850614b259122930fde49b864beafbc6966cad0e9b21f5b
+size 2289306368

modules.json

@@ -0,0 +1,23 @@
+[
+  {
+    "idx": 0,
+    "name": "transformer",
+    "path": "",
+    "type": "custom_st.Transformer",
+    "kwargs": [
+      "task"
+    ]
+  },
+  {
+    "idx": 1,
+    "name": "pooler",
+    "path": "1_Pooling",
+    "type": "sentence_transformers.models.Pooling"
+  },
+  {
+    "idx": 2,
+    "name": "normalizer",
+    "path": "2_Normalize",
+    "type": "sentence_transformers.models.Normalize"
+  }
+]

sentence_bert_config.json

@@ -0,0 +1,4 @@
+{
+  "max_seq_length": 8194,
+  "do_lower_case": false
+}

special_tokens_map.json

@@ -0,0 +1,51 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "cls_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "mask_token": {
+    "content": "<mask>",
+    "lstrip": true,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<pad>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "sep_token": {
+    "content": "</s>",
+    "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 sha256:3a56def25aa40facc030ea8b0b87f3688e4b3c39eb8b45d5702b3a1300fe2a20
+size 17082734

tokenizer_config.json

@@ -0,0 +1,55 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<pad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "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
+    },
+    "250001": {
+      "content": "<mask>",
+      "lstrip": true,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": true,
+  "cls_token": "<s>",
+  "eos_token": "</s>",
+  "extra_special_tokens": {},
+  "mask_token": "<mask>",
+  "model_max_length": 8194,
+  "pad_token": "<pad>",
+  "sep_token": "</s>",
+  "tokenizer_class": "XLMRobertaTokenizer",
+  "unk_token": "<unk>"
+}